From e3533ffea3e24ab0c253753d9ecfd7a2e6461e7c Mon Sep 17 00:00:00 2001
From: Anthony Chang <27950904+rosenrodt@users.noreply.github.com>
Date: Fri, 13 Jun 2025 09:44:04 +0800
Subject: [PATCH] update moe fp4 cubins and runner

update test_moe fp4 cases

Signed-off-by: Anthony Chang <27950904+rosenrodt@users.noreply.github.com>
---
 .../batchedGemm/KernelRunner.cpp              |  144 +-
 .../batchedGemm/KernelRunner.h                |    9 +-
 .../trtllmGen_bmm_export/BatchedGemmEnums.h   |   67 +
 .../BatchedGemmInterface.h                    |   49 +-
 .../trtllmGen_bmm_export/BatchedGemmOptions.h |  144 +-
 .../batchedGemm/trtllmGen_bmm_export/Enums.h  |   56 +
 .../GemmGatedActOptions.h                     |   30 +-
 .../trtllmGen_bmm_export/GemmOptions.h        |  287 +-
 .../trtllmGen_bmm_export/KernelMetaInfo.h     | 3266 +++++++++++------
 .../trtllmGen_bmm_export/KernelParams.h       |  422 ++-
 .../trtllmGen_bmm_export/KernelTraits.h       |   60 +-
 .../trtllmGen_bmm_export/TmaDescriptor.h      |  128 +-
 .../trtllmGen_bmm_export/config.json          |  215 +-
 ...eMmaOutput_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...persistent_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...eMmaOutput_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...persistent_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...eMmaOutput_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...transposeMmaOutput_batchN_sm100a_cubin.cpp |    3 +
 ...persistent_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...eMmaOutput_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...transposeMmaOutput_batchN_sm100a_cubin.cpp |    3 +
 ...persistent_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...tput_dsFp8_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...persistent_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...tput_dsFp8_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...persistent_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...tput_dsFp8_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...persistent_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...tput_dsFp8_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...persistent_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...tput_dsFp8_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...persistent_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...tput_dsFp8_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...persistent_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...tput_dsFp8_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...persistent_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...tput_dsFp8_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...persistent_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...tput_dsFp8_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...oseMmaOutput_dsFp8_batchN_sm100a_cubin.cpp |    3 +
 ...persistent_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...tput_dsFp8_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...oseMmaOutput_dsFp8_batchN_sm100a_cubin.cpp |    3 +
 ...persistent_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...eMmaOutput_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...transposeMmaOutput_batchN_sm100a_cubin.cpp |    3 +
 ...persistent_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...eMmaOutput_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...transposeMmaOutput_batchN_sm100a_cubin.cpp |    3 +
 ...persistent_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...eMmaOutput_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...persistent_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...eMmaOutput_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...persistent_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...eMmaOutput_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...persistent_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...eMmaOutput_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ...persistent_batchN_dynamic_sm100a_cubin.cpp |    3 +
 ..._routeLdgsts_silu_dynamic_sm100a_cubin.cpp |    3 +
 ...transposeMmaOutput_batchN_sm100a_cubin.cpp |    3 +
 ..._routeLdgsts_silu_dynamic_sm100a_cubin.cpp |    3 +
 ...transposeMmaOutput_batchN_sm100a_cubin.cpp |    3 +
 ...atchN_routeLdgsts_dynamic_sm100a_cubin.cpp |    3 +
 ...atchN_routeLdgsts_dynamic_sm100a_cubin.cpp |    3 +
 ...atchN_routeLdgsts_dynamic_sm100a_cubin.cpp |    3 +
 ...atchN_routeLdgsts_dynamic_sm100a_cubin.cpp |    3 +
 ...atchN_routeLdgsts_dynamic_sm100a_cubin.cpp |    3 +
 ...atchN_routeLdgsts_dynamic_sm100a_cubin.cpp |    3 +
 ...atchN_routeLdgsts_dynamic_sm100a_cubin.cpp |    3 +
 ...atchN_routeLdgsts_dynamic_sm100a_cubin.cpp |    3 +
 ...atchN_routeLdgsts_dynamic_sm100a_cubin.cpp |    3 +
 ...oseMmaOutput_dsFp8_batchN_sm100a_cubin.cpp |    3 +
 ...atchN_routeLdgsts_dynamic_sm100a_cubin.cpp |    3 +
 ...oseMmaOutput_dsFp8_batchN_sm100a_cubin.cpp |    3 +
 ..._routeLdgsts_silu_dynamic_sm100a_cubin.cpp |    3 +
 ...transposeMmaOutput_batchN_sm100a_cubin.cpp |    3 +
 ..._routeLdgsts_silu_dynamic_sm100a_cubin.cpp |    3 +
 ...transposeMmaOutput_batchN_sm100a_cubin.cpp |    3 +
 ...transposeMmaOutput_batchN_sm100a_cubin.cpp |    3 +
 ...transposeMmaOutput_batchN_sm100a_cubin.cpp |    3 +
 ...oseMmaOutput_dsFp8_batchN_sm100a_cubin.cpp |    3 +
 ...oseMmaOutput_dsFp8_batchN_sm100a_cubin.cpp |    3 +
 ...transposeMmaOutput_batchN_sm100a_cubin.cpp |    3 +
 ...transposeMmaOutput_batchN_sm100a_cubin.cpp |    3 +
 ..._routeLdgsts_silu_dynamic_sm100a_cubin.cpp |    3 +
 ..._routeLdgsts_silu_dynamic_sm100a_cubin.cpp |    3 +
 tests/unittest/_torch/thop/test_moe.py        |   10 +-
 88 files changed, 3650 insertions(+), 1459 deletions(-)
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/BatchedGemmEnums.h
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x128x128_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x128x128_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x128x128u2_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x128x128u2_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x16x128_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x16x128_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x16x128u2_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x16x128u2_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x32x128_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x32x128_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x32x128u2_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x32x128u2_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x64x128_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x64x128_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x64x128u2_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x64x128u2_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x256u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x256u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x128x128_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x128x128u2_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x16x128_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x16x128u2_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x32x128_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x32x128u2_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x64x128_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x64x128u2_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Fp16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Fp16_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Fp16_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Fp16_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Fp16_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Fp16_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_MxE4m3_MxE2m1MxE4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin.cpp
 create mode 100644 cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_MxE4m3_MxE2m1MxE4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin.cpp

diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/KernelRunner.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/KernelRunner.cpp
index 6f6e186d71..de5f1f650d 100644
--- a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/KernelRunner.cpp
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/KernelRunner.cpp
@@ -27,6 +27,138 @@ namespace kernels
 {
 
 using namespace batchedGemm::batchedGemm;
+using namespace batchedGemm::gemm;
+using namespace batchedGemm::trtllm::gen;
+
+std::vector<int64_t> prioritizePredefinedConfigs(int m, int n, int k, std::vector<int64_t> const& sortedIndices,
+    batchedGemm::batchedGemm::BatchedGemmConfig const* configs)
+{
+
+    // Function to bubble up the pre-determined config.
+    auto bubbleUpConfig = [&configs](std::vector<int64_t> const& sortedIndices, auto&& pred) -> std::vector<int64_t>
+    {
+        std::vector<int64_t> prioritizedIndices_;
+        // Copy matching configs to new vector
+        std::copy_if(sortedIndices.begin(), sortedIndices.end(), std::back_inserter(prioritizedIndices_),
+            [&configs, &pred](int idx)
+            {
+                BatchedGemmConfig const& config = configs[idx];
+                return (pred(config));
+            });
+        // Copy the rest of the configs to new vector, if not already copied
+        std::copy_if(sortedIndices.begin(), sortedIndices.end(), std::back_inserter(prioritizedIndices_),
+            [&prioritizedIndices_](int idx) {
+                return std::find(prioritizedIndices_.begin(), prioritizedIndices_.end(), idx)
+                    == prioritizedIndices_.end();
+            });
+        return prioritizedIndices_;
+    };
+
+    // Init empty vector
+    std::vector<int64_t> prioritizedIndices;
+
+    //
+    // Qwen3
+    //
+
+    // Qwen3_235B_TP1_EP8_MoE_FC1 m=3072 k=4096
+    if (n /* out_dim */ == 3072 && k /* in_dim */ == 4096)
+    {
+        auto pred = [](BatchedGemmConfig const& config)
+        {
+            BatchedGemmOptions const& options = config.mOptions;
+            return options.mNumStages == 4 && options.mNumStagesMma == 1 && options.mTileK == 512
+                && options.mTileScheduler == TileScheduler::Static;
+        };
+        prioritizedIndices = bubbleUpConfig(sortedIndices, pred);
+    }
+    // Qwen3_235B_TP1_EP8_MoE_FC2 m=4096 k=1536
+    else if (n /* out_dim */ == 4096 && k /* in_dim */ == 1536)
+    {
+        auto pred = [](BatchedGemmConfig const& config)
+        {
+            BatchedGemmOptions const& options = config.mOptions;
+            return options.mNumStages == 4 && options.mNumStagesMma == 1 && options.mTileK == 512
+                && options.mTileScheduler == TileScheduler::Static;
+        };
+        prioritizedIndices = bubbleUpConfig(sortedIndices, pred);
+    }
+    // Qwen3_235B_TP2_EP4_MoE_FC1 m=1536 k=4096
+    else if (n /* out_dim */ == 1536 && k /* in_dim */ == 4096)
+    {
+        auto pred = [](BatchedGemmConfig const& config)
+        {
+            BatchedGemmOptions const& options = config.mOptions;
+            return options.mNumStages == 4 && options.mNumStagesMma == 1 && options.mTileK == 512
+                && options.mTileScheduler == TileScheduler::Static;
+        };
+        prioritizedIndices = bubbleUpConfig(sortedIndices, pred);
+    }
+    // Qwen3_235B_TP2_EP4_MoE_FC2 m=4096 k=768
+    else if (n /* out_dim */ == 4096 && k /* in_dim */ == 768)
+    {
+        auto pred = [](BatchedGemmConfig const& config)
+        {
+            BatchedGemmOptions const& options = config.mOptions;
+            return options.mNumStages == 4 && options.mNumStagesMma == 2 && options.mTileK == 512
+                && options.mTileScheduler == TileScheduler::Persistent;
+        };
+        prioritizedIndices = bubbleUpConfig(sortedIndices, pred);
+    }
+    // Qwen3_235B_TP4_EP2_MoE_FC1 m=768 k=4096
+    else if (n /* out_dim */ == 768 && k /* in_dim */ == 4096)
+    {
+        auto pred = [](BatchedGemmConfig const& config)
+        {
+            BatchedGemmOptions const& options = config.mOptions;
+            return options.mNumStages == 4 && options.mNumStagesMma == 1 && options.mTileK == 512
+                && options.mTileScheduler == TileScheduler::Static;
+        };
+        prioritizedIndices = bubbleUpConfig(sortedIndices, pred);
+    }
+    // Qwen3_235B_TP4_EP2_MoE_FC2 m=4096 k=384
+    else if (n /* out_dim */ == 4096 && k /* in_dim */ == 384)
+    {
+        auto pred = [](BatchedGemmConfig const& config)
+        {
+            BatchedGemmOptions const& options = config.mOptions;
+            return options.mNumStages == 4 && options.mNumStagesMma == 2 && options.mTileK == 512
+                && options.mTileScheduler == TileScheduler::Persistent;
+        };
+        prioritizedIndices = bubbleUpConfig(sortedIndices, pred);
+    }
+    // Qwen3_235B_TP8_EP1_MoE_FC1 m=384 k=4096
+    else if (n /* out_dim */ == 384 && k /* in_dim */ == 4096)
+    {
+        auto pred = [](BatchedGemmConfig const& config)
+        {
+            BatchedGemmOptions const& options = config.mOptions;
+            return options.mNumStages == 4 && options.mNumStagesMma == 1 && options.mTileK == 512
+                && options.mTileScheduler == TileScheduler::Static;
+        };
+        prioritizedIndices = bubbleUpConfig(sortedIndices, pred);
+    }
+    // Qwen3_235B_TP8_EP1_MoE_FC2 m=4096 k=192
+    else if (n /* out_dim */ == 4096 && k /* in_dim */ == 192)
+    {
+        auto pred = [](BatchedGemmConfig const& config)
+        {
+            BatchedGemmOptions const& options = config.mOptions;
+            return options.mNumStages == 4 && options.mNumStagesMma == 2 && options.mTileK == 256
+                && options.mTileScheduler == TileScheduler::Persistent;
+        };
+        prioritizedIndices = bubbleUpConfig(sortedIndices, pred);
+    }
+    //
+    // Fall back
+    //
+    else
+    {
+        prioritizedIndices = sortedIndices;
+    }
+
+    return prioritizedIndices;
+}
 
 TrtllmGenBatchedGemmRunner::TrtllmGenBatchedGemmRunner(TrtllmGenBatchedGemmRunnerOptions const& options_)
     : mOptions(options_)
@@ -44,7 +176,8 @@ TrtllmGenBatchedGemmRunner::TrtllmGenBatchedGemmRunner(TrtllmGenBatchedGemmRunne
         // When we include low-latency kernels we can set transposeMmaOutput via constructor
         if (options.mDtypeA == mOptions.eltType && options.mDtypeC == mOptions.outputType
             && options.mUseDeepSeekFp8 == mOptions.deepSeekFp8
-            && options.mTransposeMmaOutput == mOptions.transposeMmaOutput && options.mRouteAct == mOptions.routeAct
+            && options.mTransposeMmaOutput == mOptions.transposeMmaOutput
+            && (!doesRouteImplUseNoRoute(options.mRouteImpl)) == mOptions.routeAct
             && options.mFusedAct == mOptions.fusedAct && options.mIsStaticBatch == mOptions.staticBatch
             && tileSize == mOptions.tileSize)
         {
@@ -227,9 +360,9 @@ std::vector<int64_t> TrtllmGenBatchedGemmRunner::getValidConfigIndices(int32_t m
     gemmData.mProblemDimensions.mWorldSize = 1;
     gemmData.mProblemDimensions.mMaxNumCtasInTokenDim = maxNumCtasInBatchDim;
     // Sort configs by options
-    std::vector<int32_t> sortedIndices = mPassingConfigIndices;
+    std::vector<int64_t> sortedIndices = mPassingConfigIndices;
     std::sort(sortedIndices.begin(), sortedIndices.end(),
-        [&configs](int32_t idx0, int32_t idx1)
+        [&configs](int64_t idx0, int64_t idx1)
         {
             auto const& optionsA = configs[idx0].mOptions;
             auto const& optionsB = configs[idx1].mOptions;
@@ -247,7 +380,7 @@ std::vector<int64_t> TrtllmGenBatchedGemmRunner::getValidConfigIndices(int32_t m
             }
 
             // Then by tile scheduler (persistent scheduler is better for FC2 in MoE)
-            if (!optionsA.mRouteAct)
+            if (doesRouteImplUseNoRoute(optionsA.mRouteImpl))
             {
                 return optionsA.mTileScheduler == batchedGemm::gemm::TileScheduler::Persistent;
             }
@@ -255,8 +388,9 @@ std::vector<int64_t> TrtllmGenBatchedGemmRunner::getValidConfigIndices(int32_t m
             return optionsA.mTileM > optionsB.mTileM;
         });
 
+    std::vector<int64_t> prioritizedIndices = prioritizePredefinedConfigs(m, n, k, sortedIndices, configs);
     std::vector<int64_t> validConfigIndices;
-    for (auto const& configIndex : sortedIndices)
+    for (auto const& configIndex : prioritizedIndices)
     {
         auto const& config = configs[configIndex];
         auto isValidConfig = bmm.isValidConfig(config, gemmData);
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/KernelRunner.h b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/KernelRunner.h
index 0584bc0bb4..7fe892511b 100644
--- a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/KernelRunner.h
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/KernelRunner.h
@@ -50,6 +50,7 @@ class TrtllmGenBatchedGemmRunner
         std::vector<int32_t> const& batchedTokens, int32_t numTokens, int32_t numBatches, int32_t maxNumCtasInBatchDim,
         std::optional<int32_t> configIndex = std::nullopt);
 
+    // Generic GEMM interface
     void run(int32_t m, int32_t n, int32_t k, std::vector<int32_t> const& batchedTokens, int32_t numTokens,
         int32_t numBatches, int32_t maxNumCtasInBatchDim, void const* a, void const* sfA, void const* b,
         void const* sfB, void const* perTokensSfA, void const* perTokensSfB, float const* scaleC,
@@ -57,16 +58,18 @@ class TrtllmGenBatchedGemmRunner
         int32_t const* ctaIdxXyToBatchIdx, int32_t const* ctaIdxXyToMnLimit, int32_t const* numNonExitingCtas,
         void* workspace, CUstream stream, int device, std::optional<int32_t> configIndex = std::nullopt);
 
+    // NVFP4 per-block scaling GEMM
     void run(int32_t m, int32_t n, int32_t k, std::vector<int32_t> const& batchedTokens, void const* a, void const* sfA,
         void const* b, void const* sfB, void* c, void* outSfC, void* workspace, CUstream stream, int device,
         std::optional<int32_t> configIndex = std::nullopt);
 
+    // FP8 per-tensor scaling GEMM
     void run(int32_t m, int32_t n, int32_t k, std::vector<int32_t> const& batchedTokens, void const* a, void const* b,
         float const* scaleC, float const* scaleGateC, void* c, void* workspace, CUstream stream, int device,
         std::optional<int32_t> configIndex = std::nullopt);
 
     // Get the list of configs that passed the validation based on the constructor options
-    [[nodiscard]] std::vector<int32_t> getPassingConfigIndices() const
+    [[nodiscard]] std::vector<int64_t> getPassingConfigIndices() const
     {
         return mPassingConfigIndices;
     }
@@ -88,8 +91,8 @@ class TrtllmGenBatchedGemmRunner
 
 private:
     TrtllmGenBatchedGemmRunnerOptions mOptions;
-    std::vector<int32_t> mPassingConfigIndices;
-    std::optional<int32_t> mSelectedConfigIndex;
+    std::vector<int64_t> mPassingConfigIndices;
+    std::optional<int64_t> mSelectedConfigIndex;
 };
 } // namespace kernels
 } // namespace tensorrt_llm
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/BatchedGemmEnums.h b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/BatchedGemmEnums.h
new file mode 100644
index 0000000000..ffea777396
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/BatchedGemmEnums.h
@@ -0,0 +1,67 @@
+/*
+ * SPDX-FileCopyrightText: Copyright (c) 1993-2025 NVIDIA CORPORATION &
+ * AFFILIATES. All rights reserved. SPDX-License-Identifier: Apache-2.0
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#pragma once
+
+#include <cassert>
+#include <string>
+
+namespace batchedGemm
+{
+
+namespace batchedGemm
+{
+
+////////////////////////////////////////////////////////////////////////////////////////////////////
+
+enum class RouteImpl
+{
+    // No Routing
+    NoRoute = 0,
+    // Use LDGSTS to do the routing
+    Ldgsts = 1,
+    // Use UTMALDG.GATHER4 to do the routing
+    Tma = 2
+};
+
+////////////////////////////////////////////////////////////////////////////////////////////////////
+
+inline bool doesRouteImplUseNoRoute(RouteImpl mode)
+{
+    return (mode == RouteImpl::NoRoute);
+}
+
+////////////////////////////////////////////////////////////////////////////////////////////////////
+
+inline bool doesRouteImplUseLdgsts(RouteImpl mode)
+{
+    return (mode == RouteImpl::Ldgsts);
+}
+
+////////////////////////////////////////////////////////////////////////////////////////////////////
+
+inline bool doesRouteImplUseTma(RouteImpl mode)
+{
+    return (mode == RouteImpl::Tma);
+}
+
+////////////////////////////////////////////////////////////////////////////////////////////////////
+
+} // namespace batchedGemm
+
+////////////////////////////////////////////////////////////////////////////////////////////////////
+
+} // namespace batchedGemm
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/BatchedGemmInterface.h b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/BatchedGemmInterface.h
index e2de17c6de..4251a333e0 100644
--- a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/BatchedGemmInterface.h
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/BatchedGemmInterface.h
@@ -96,12 +96,16 @@ struct BatchedGemmData
         //      Logical strides are [K, 1].
         //
         //   If batchN:
-        //      If transposeMatrixA is false
+        //      If layoutA is MatrixLayout::MajorK
         //         Logical shape is [B, divUpMul(M, tileM), K].
         //         Logical strides are [divUpMul(M, tileM) * K, K, 1].
-        //      If transposeMatrixA is true
+        //      If layoutA is MatrixLayout::MajorMn
         //         Logical shape is [B, K, divUpMul(M, tileM)].
         //         Logical strides are [K * divUpMul(M, tileM), divUpMul(M, tileM), 1].
+        //      If layoutA is MatrixLayout::BlockMajorK
+        //         Logical shape is [B, K / blockK, divUpMul(M, tileM), blockK].
+        //         Logical strides are [K * divUpMul(M, tileM), divUpMul(M, tileM) * blockK, blockK, 1].
+        //         where blockK is 128B.
         void const* mPtrA{nullptr};
 
         // The block scaling factors to dequantize A.
@@ -154,12 +158,16 @@ struct BatchedGemmData
         //      Logical strides are [K, 1].
         //
         //   If batchM:
-        //      If transposeMatrixB is true
+        //      If layoutB is MatrixLayout::MajorK
         //         Logical shape is [B, divUpMul(N, tileN), K].
         //         Logical strides are [divUpMul(N, tileN) * K, K, 1].
-        //      If transposeMatrixB is false
+        //      If layoutB is MatrixLayout::MajorMn
         //         Logical shape is [B, K, divUpMul(N, tileN)].
         //         Logical strides are [K * divUpMul(N, tileN), divUpMul(N, tileN), 1].
+        //      If layoutB is MatrixLayout::BlockMajorK
+        //         Logical shape is [B, K / blockK, divUpMul(N, tileN), blockK].
+        //         Logical strides are [K * divUpMul(N, tileN), divUpMul(N, tileN) * blockK, blockK, 1].
+        //         where blockK is 128B.
         void const* mPtrB{nullptr};
 
         // The scaling factors to dequantize B.
@@ -210,6 +218,21 @@ struct BatchedGemmData
         //     Logical shape is [sum(divUpMul(N[bi], tileN) for bi in B)]
         void const* mPtrPerTokenSfB{nullptr};
 
+        // The bias applied after the GEMM and before the activation function.
+        // The bias is applied before applying the global scaling factor. I.e.
+        // C = act(A * B + bias') * scaleC
+        // scaleC = dequantA * dequantB * quantC
+        // Thus, the bias' = bias / (dequantA * dequantB), where the bias is the original bias.
+        //
+        // If batchM, BiasType must be N, and bias shape is [B, N].
+        // The bias is broadcasted along the M dimension.
+        //
+        // If batchN BiasType must be M, and bias shape is [B, M].
+        // The bias is broadcasted along the N dimension.
+        //
+        // The dtype is float32.
+        void const* mPtrBias{nullptr};
+
         // The output tensor scaling factor for MxFp{4,8}, Fp8 and NvFp4 quantization.
         // TensorRT-LLM API requires a scaling factor on the device.
         // Shape is [B].
@@ -220,6 +243,12 @@ struct BatchedGemmData
         // Shape is [B].
         float const* mPtrScaleGate{nullptr};
 
+        // The alpha and beta for SwiGlu.
+        // gatedActivation <- (x0 + beta) * sigmoid(alpha * x1)
+        // Shape is [B]
+        float const* mPtrSwiGluAlpha{nullptr};
+        float const* mPtrSwiGluBeta{nullptr};
+
         // Param is used when the kernel is configured with -routeAct true.
         // The inputs are not padded, but the outputs are padded to divUpMul(M[bi], tileM) for batchM or
         // divUpMul(N[bi], tileN) for batchN.
@@ -609,11 +638,13 @@ int32_t BatchedGemmInterface::run(BatchedGemmConfig const& config, void* workspa
         batchedGemmData.mInputBuffers.mPtrB, batchedGemmData.mOutputBuffers.mPtrC,
         batchedGemmData.mInputBuffers.mPtrSfA, batchedGemmData.mInputBuffers.mPtrSfB,
         batchedGemmData.mInputBuffers.mPtrPerTokenSfA, batchedGemmData.mInputBuffers.mPtrPerTokenSfB,
-        batchedGemmData.mOutputBuffers.mPtrSfC, batchedGemmData.mInputBuffers.mPtrScaleC,
-        batchedGemmData.mInputBuffers.mPtrScaleGate, batchedGemmData.mInputBuffers.mPtrRouteMap, dPtrRowMax,
-        dPtrRowMaxBars, batchedGemmData.mInputBuffers.mPtrNumNonExitingCtas,
-        batchedGemmData.mInputBuffers.mPtrTotalNumPaddedTokens, batchedGemmData.mInputBuffers.mPtrCtaIdxXyToBatchIdx,
-        batchedGemmData.mInputBuffers.mPtrCtaIdxXyToMnLimit, maxNumCtasInBatchDim);
+        batchedGemmData.mInputBuffers.mPtrBias, batchedGemmData.mOutputBuffers.mPtrSfC,
+        batchedGemmData.mInputBuffers.mPtrScaleC, batchedGemmData.mInputBuffers.mPtrScaleGate,
+        batchedGemmData.mInputBuffers.mPtrSwiGluAlpha, batchedGemmData.mInputBuffers.mPtrSwiGluBeta,
+        batchedGemmData.mInputBuffers.mPtrRouteMap, dPtrRowMax, dPtrRowMaxBars,
+        batchedGemmData.mInputBuffers.mPtrNumNonExitingCtas, batchedGemmData.mInputBuffers.mPtrTotalNumPaddedTokens,
+        batchedGemmData.mInputBuffers.mPtrCtaIdxXyToBatchIdx, batchedGemmData.mInputBuffers.mPtrCtaIdxXyToMnLimit,
+        maxNumCtasInBatchDim);
 
     // The size of the grid.
     std::vector<int32_t> grid{numCtaX, numCtaY, numCtaZ};
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/BatchedGemmOptions.h b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/BatchedGemmOptions.h
index f59cdd1fff..8baf99a9f8 100644
--- a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/BatchedGemmOptions.h
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/BatchedGemmOptions.h
@@ -16,6 +16,7 @@
  */
 #pragma once
 
+#include "BatchedGemmEnums.h"
 #include "GemmGatedActOptions.h"
 #include "GemmOptions.h"
 
@@ -81,33 +82,36 @@ struct BatchedGemmOptions : public gemmGatedAct::GemmGatedActOptions
     BatchedGemmOptions() = default;
 
     // FIXME We create explicit constructor with all options to WAR stubgen issue in TRT-LLM.
-    BatchedGemmOptions(gemm::AllReduceAlgo allReduceAlgo, int clusterDimX, int clusterDimY, int clusterDimZ,
-        tg::Dtype dtypeAcc, tg::Dtype dtypeA, tg::Dtype dtypeB, tg::Dtype dtypeC, bool enablesEarlyExit,
-        bool enablesDelayedEarlyExit, bool enablesGlobalPtxKnobs, int epilogueLdtmDps, int epilogueLdtmBits,
-        int epilogueTileM, int epilogueTileN, bool gridTriggerSecondaryA, bool gridTriggerSecondaryB,
-        bool gridWaitForPrimaryEarlyExit, bool gridWaitForPrimaryA, bool gridWaitForPrimaryB, bool hoistLoadTaskInit,
-        bool hoistMmaTaskTryWaits, int k, gemm::KernelTraits kernelTraits, int m, int mmaK, tg::MmaKind mmaKind,
-        int mmaM, int mmaN, bool mockAllReduce, int n, int numSlicesForSplitK, int numSlicesForSliceK, int numStages,
-        int numStagesMma, int numStagesMmaWithinWorkTile, int numStagesMmaAcrossWorkTile, int numStagesWorkId,
-        bool outputDebugTensors, bool useShuffledMatrixA, bool sliceK, gemm::SplitK splitK, bool transposeMatrixA,
-        bool transposeMatrixB, bool transposeMmaOutput, int tileM, int tileN, int tileK, bool useUnrollLoop2xForMma,
-        bool useCustomMmaSchedule, bool useHoistTryWaitForCustomMmaSchedule, bool useDeepSeekFp8, bool usePerTokenSfA,
-        bool usePerTokenSfB, bool useTmaStore, bool useTwoTmaLoadWarps, bool useTwoMmaWarps, tg::SfLayout sfLayoutA,
-        tg::SfLayout sfLayoutB, tg::SfLayout sfLayoutC, gemm::TileScheduler tileScheduler,
-        gemmGatedAct::ActType actType, std::vector<int> batchedM, std::vector<int> batchedN, BatchMode batchMode,
-        int numBatches, bool isStaticBatch, int numTokens, bool routeAct, bool gridWaitForPrimaryRouting, bool fusedAct,
-        int numRegsPerThreadNonEpilogueWarp, int numRegsPerThreadEpilogueWarp)
+    BatchedGemmOptions(gemm::AllReduceAlgo allReduceAlgo, gemm::BiasType biasType, int blockK, int clusterDimX,
+        int clusterDimY, int clusterDimZ, tg::Dtype dtypeAcc, tg::Dtype dtypeA, tg::Dtype dtypeB, tg::Dtype dtypeC,
+        tg::Dtype dtypeMmaA, tg::Dtype dtypeMmaB, bool enablesEarlyExit, bool enablesDelayedEarlyExit,
+        bool enablesGlobalPtxKnobs, int epilogueLdtmDps, int epilogueLdtmBits, int epilogueTileM, int epilogueTileN,
+        bool gridTriggerSecondaryA, bool gridTriggerSecondaryB, bool gridWaitForPrimaryEarlyExit,
+        bool gridWaitForPrimaryA, bool gridWaitForPrimaryB, bool hoistLoadTaskInit, bool hoistMmaTaskTryWaits, int k,
+        gemm::KernelTraits kernelTraits, gemm::MatrixLayout layoutA, gemm::MatrixLayout layoutB, int m, int mmaK,
+        tg::MmaKind mmaKind, int mmaM, int mmaN, bool mockAllReduce, int n, int numSlicesForSplitK,
+        int numSlicesForSliceK, int numStages, int numStagesMma, int numStagesMmaWithinWorkTile,
+        int numStagesMmaAcrossWorkTile, int numStagesWorkId, bool outputDebugTensors, bool useShuffledMatrixA,
+        bool sliceK, gemm::SplitK splitK, bool transposeMmaOutput, int tileM, int tileN, int tileK,
+        bool useUnrollLoop2xForMma, bool useCustomMmaSchedule, bool useHoistTryWaitForCustomMmaSchedule,
+        bool useDeepSeekFp8, bool usePerTokenSfA, bool usePerTokenSfB, bool useTmaStore, bool useTwoTmaLoadWarps,
+        bool useTwoMmaWarps, tg::SfLayout sfLayoutA, tg::SfLayout sfLayoutB, tg::SfLayout sfLayoutC,
+        int32_t sfReshapeFactor, gemm::TileScheduler tileScheduler, gemmGatedAct::ActType actType,
+        std::vector<int> batchedM, std::vector<int> batchedN, BatchMode batchMode, int numBatches, bool isStaticBatch,
+        int numTokens, RouteImpl routeImpl, bool gridWaitForPrimaryRouting, bool fusedAct,
+        int numRegsPerThreadNonEpilogueWarp, int numRegsPerThreadEpilogueWarp, int numRegsCastAWarps)
         : gemmGatedAct::GemmGatedActOptions(
-            gemm::GemmOptions(allReduceAlgo, clusterDimX, clusterDimY, clusterDimZ, dtypeAcc, dtypeA, dtypeB, dtypeC,
-                enablesEarlyExit, enablesDelayedEarlyExit, enablesGlobalPtxKnobs, epilogueLdtmDps, epilogueLdtmBits,
-                epilogueTileM, epilogueTileN, gridTriggerSecondaryA, gridTriggerSecondaryB, gridWaitForPrimaryEarlyExit,
-                gridWaitForPrimaryA, gridWaitForPrimaryB, hoistLoadTaskInit, hoistMmaTaskTryWaits, k, kernelTraits, m,
-                mmaK, mmaKind, mmaM, mmaN, mockAllReduce, n, numSlicesForSplitK, numSlicesForSliceK, numStages,
-                numStagesMma, numStagesMmaWithinWorkTile, numStagesMmaAcrossWorkTile, numStagesWorkId,
-                outputDebugTensors, useShuffledMatrixA, sliceK, splitK, transposeMatrixA, transposeMatrixB,
-                transposeMmaOutput, tileM, tileN, tileK, useUnrollLoop2xForMma, useCustomMmaSchedule,
-                useHoistTryWaitForCustomMmaSchedule, useDeepSeekFp8, usePerTokenSfA, usePerTokenSfB, useTmaStore,
-                useTwoTmaLoadWarps, useTwoMmaWarps, sfLayoutA, sfLayoutB, sfLayoutC, tileScheduler),
+            gemm::GemmOptions(allReduceAlgo, biasType, blockK, clusterDimX, clusterDimY, clusterDimZ, dtypeAcc, dtypeA,
+                dtypeB, dtypeC, dtypeMmaA, dtypeMmaB, enablesEarlyExit, enablesDelayedEarlyExit, enablesGlobalPtxKnobs,
+                epilogueLdtmDps, epilogueLdtmBits, epilogueTileM, epilogueTileN, gridTriggerSecondaryA,
+                gridTriggerSecondaryB, gridWaitForPrimaryEarlyExit, gridWaitForPrimaryA, gridWaitForPrimaryB,
+                hoistLoadTaskInit, hoistMmaTaskTryWaits, k, kernelTraits, layoutA, layoutB, m, mmaK, mmaKind, mmaM,
+                mmaN, mockAllReduce, n, numSlicesForSplitK, numSlicesForSliceK, numStages, numStagesMma,
+                numStagesMmaWithinWorkTile, numStagesMmaAcrossWorkTile, numStagesWorkId, outputDebugTensors,
+                useShuffledMatrixA, sliceK, splitK, transposeMmaOutput, tileM, tileN, tileK, useUnrollLoop2xForMma,
+                useCustomMmaSchedule, useHoistTryWaitForCustomMmaSchedule, useDeepSeekFp8, usePerTokenSfA,
+                usePerTokenSfB, useTmaStore, useTwoTmaLoadWarps, useTwoMmaWarps, sfLayoutA, sfLayoutB, sfLayoutC,
+                sfReshapeFactor, tileScheduler),
             actType)
         , mBatchedM(batchedM)
         , mBatchedN(batchedN)
@@ -115,11 +119,12 @@ struct BatchedGemmOptions : public gemmGatedAct::GemmGatedActOptions
         , mNumBatches(numBatches)
         , mIsStaticBatch(isStaticBatch)
         , mNumTokens(numTokens)
-        , mRouteAct(routeAct)
+        , mRouteImpl(routeImpl)
         , mGridWaitForPrimaryRouting(gridWaitForPrimaryRouting)
         , mFusedAct(fusedAct)
         , mNumRegsPerThreadNonEpilogueWarp(numRegsPerThreadNonEpilogueWarp)
         , mNumRegsPerThreadEpilogueWarp(numRegsPerThreadEpilogueWarp)
+        , mNumRegsCastAWarps(numRegsCastAWarps)
     {
     }
 
@@ -137,7 +142,7 @@ struct BatchedGemmOptions : public gemmGatedAct::GemmGatedActOptions
     // Total number of tokens.
     int mNumTokens{32};
     // Whether load the input tokens and do routing.
-    bool mRouteAct{false};
+    RouteImpl mRouteImpl{RouteImpl::NoRoute};
     // Whether the loads that load from ptrRouteMap, ptrTotalNumPaddedTokens,
     // ptrCtaIdxXyToBatchIdx, etc.. should wait on a grid dependency.
     bool mGridWaitForPrimaryRouting{true};
@@ -149,6 +154,8 @@ struct BatchedGemmOptions : public gemmGatedAct::GemmGatedActOptions
     int mNumRegsPerThreadNonEpilogueWarp{0};
     // Number of registers per thread for epilogue warps
     int mNumRegsPerThreadEpilogueWarp{0};
+    // Number of registers for the cast A warps.
+    int mNumRegsCastAWarps{0};
 };
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////
@@ -196,19 +203,15 @@ bool checkAndUpdateBatchedGemmOptions(BatchedGemmOptions& options, bool isBlackw
         if (batchM)
         {
             TLLM_CHECK_ERROR(options.mN > 0 && options.mK > 0, "N and K must be larger than 0");
-            TLLM_CHECK_ERROR(options.mN >= options.mTileN && options.mK >= options.mTileK,
-                "N and K must be equal or larger than TileN and TileK respectively.");
-            TLLM_CHECK_ERROR(options.mN % options.mTileN == 0 && options.mK % options.mTileK == 0,
-                "N and K must be divisible by TileN and TileK respectively.");
+            TLLM_CHECK_ERROR(options.mN >= options.mTileN, "N must be equal or larger than TileN.");
+            TLLM_CHECK_ERROR(options.mN % options.mTileN == 0, "N must be divisible by TileN.");
             TLLM_CHECK_ERROR(!options.mTransposeMmaOutput, "When batchM the MMA output has to be in row-major.");
         }
         else
         {
             TLLM_CHECK_ERROR(options.mM > 0 && options.mK > 0, "M and K must be larger than 0");
-            TLLM_CHECK_ERROR(options.mM >= options.mTileM && options.mK >= options.mTileK,
-                "N and K must be equal or larger than tileN and tileK respectively.");
-            TLLM_CHECK_ERROR(options.mM % options.mTileM == 0 && options.mK % options.mTileK == 0,
-                "M and K must be divisible by TileM and TileK respectively.");
+            TLLM_CHECK_ERROR(options.mM >= options.mTileM, "N must be equal or larger than tileN.");
+            TLLM_CHECK_ERROR(options.mM % options.mTileM == 0, "M must be divisible by TileM.");
             TLLM_CHECK_ERROR(options.mTransposeMmaOutput, "When batchN the MMA output has to be in column-major.");
         }
     }
@@ -240,26 +243,82 @@ bool checkAndUpdateBatchedGemmOptions(BatchedGemmOptions& options, bool isBlackw
     {
         if (options.mDtypeA == tg::Dtype::MxE2m1 && options.mMmaKind == tg::MmaKind::MxFp8Fp6Fp4)
         {
-            TLLM_CHECK_ERROR(!options.mRouteAct, "RouteAct is not supported with dtypeA = MxE2m1 and MxFp8Fp6Fp4.");
+            TLLM_CHECK_ERROR(doesRouteImplUseNoRoute(options.mRouteImpl),
+                "RouteAct is not supported with dtypeA = MxE2m1 and MxFp8Fp6Fp4.");
         }
     }
     else
     {
         if (options.mDtypeB == tg::Dtype::MxE2m1 && options.mMmaKind == tg::MmaKind::MxFp8Fp6Fp4)
         {
-            TLLM_CHECK_ERROR(!options.mRouteAct, "RouteAct is not supported with dtypeB = MxE2m1 and MxFp8Fp6Fp4.");
+            TLLM_CHECK_ERROR(doesRouteImplUseNoRoute(options.mRouteImpl),
+                "RouteAct is not supported with dtypeB = MxE2m1 and MxFp8Fp6Fp4.");
         }
     }
 
     TLLM_CHECK_ERROR(options.mUseTmaStore, "Only TMA store is supported.");
     if (batchM)
     {
-        TLLM_CHECK_ERROR(!options.mTransposeMatrixA, "Activations must be in k-major format");
+        TLLM_CHECK_ERROR(options.mLayoutA == gemm::MatrixLayout::MajorK, "Activations must be in k-major format");
     }
     else
     {
-        TLLM_CHECK_ERROR(options.mTransposeMatrixB, "Activations must be in k-major format");
+        TLLM_CHECK_ERROR(options.mLayoutB == gemm::MatrixLayout::MajorK, "Activations must be in k-major format");
     }
+
+    if (tg::mmaKindIsBlockFmt(options.mMmaKind) && !options.mUseDeepSeekFp8)
+    {
+        if (!doesRouteImplUseNoRoute(options.mRouteImpl))
+        {
+            if (batchM)
+            {
+                TLLM_CHECK_ERROR(
+                    options.mSfLayoutA == tg::SfLayout::Linear, "Tokens need use SF linear layout when being routed");
+            }
+            else
+            {
+                TLLM_CHECK_ERROR(
+                    options.mSfLayoutB == tg::SfLayout::Linear, "Tokens need use SF linear layout when being routed");
+            }
+        }
+
+        if (doesRouteImplUseTma(options.mRouteImpl))
+        {
+            TLLM_CHECK_ERROR(!batchM, "UTMALDG.GATHER4 only supported for batch N.");
+
+            if (tg::mmaKindIsBlockFmt(options.mMmaKind))
+            {
+                auto dtypeRoute = batchM ? options.mDtypeA : options.mDtypeB;
+                TLLM_CHECK_ERROR(options.mTileK % tg::dtypeNumEltsPerSf(dtypeRoute) == 0,
+                    "tileK needs to be a multiple of 16 * tg::dtypeNumEltsPerSf(dtypeA).");
+                TLLM_CHECK_ERROR(options.mTileK % (tg::dtypeNumEltsPerSf(dtypeRoute) * 16) == 0,
+                    "tileK needs to be a multiple of 16 * tg::dtypeNumEltsPerSf(dtypeA).");
+            }
+        }
+
+        if (!batchM || doesRouteImplUseNoRoute(options.mRouteImpl))
+        {
+            TLLM_CHECK_ERROR(options.mSfLayoutA == tg::SfLayout::R128c4,
+                "options.mSfLayoutA has to be tg::SfLayout::R128c4 when not being routed");
+        }
+    }
+
+    if (!gemm::isBiasTypeNone(options.mBiasType))
+    {
+        TLLM_CHECK_ERROR(
+            (gemm::isBiasTypeN(options.mBiasType) && options.mBatchMode == BatchedGemmOptions::BatchMode::BatchM)
+                || (gemm::isBiasTypeM(options.mBiasType)
+                    && options.mBatchMode == BatchedGemmOptions::BatchMode::BatchN),
+            "BatchedGemm supports only per channel bias.");
+    }
+
+    // We do not handle the case where K is not a multiple of TileK.
+    // TMA based load handles the case transparently.
+    if (doesRouteImplUseLdgsts(options.mRouteImpl))
+    {
+        TLLM_CHECK_ERROR(options.mK % options.mTileK == 0, "K must be a multiple of TileK");
+    }
+
     return isValid;
 }
 
@@ -301,11 +360,12 @@ inline std::string dumpOptions(BatchedGemmOptions const& options)
     ss << "mNumBatches=" << options.mNumBatches << "," << std::endl;
     ss << "mIsStaticBatch=" << options.mIsStaticBatch << "," << std::endl;
     ss << "mNumTokens=" << options.mNumTokens << "," << std::endl;
-    ss << "mRouteAct=" << options.mRouteAct << "," << std::endl;
+    ss << "mRouteImpl=batchedGemm::RouteImpl(" << static_cast<int32_t>(options.mRouteImpl) << ")," << std::endl;
     ss << "mGridWaitForPrimaryRouting=" << options.mGridWaitForPrimaryRouting << "," << std::endl;
     ss << "mFusedAct=" << options.mFusedAct << "," << std::endl;
     ss << "mNumRegsPerThreadNonEpilogueWarp=" << options.mNumRegsPerThreadNonEpilogueWarp << "," << std::endl;
-    ss << "mNumRegsPerThreadEpilogueWarp=" << options.mNumRegsPerThreadEpilogueWarp << std::endl;
+    ss << "mNumRegsPerThreadEpilogueWarp=" << options.mNumRegsPerThreadEpilogueWarp << "," << std::endl;
+    ss << "mNumRegsCastAWarps=" << options.mNumRegsCastAWarps << std::endl;
     return ss.str();
 }
 
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/Enums.h b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/Enums.h
index 57cbc93982..8b517526a4 100644
--- a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/Enums.h
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/Enums.h
@@ -39,6 +39,31 @@ enum class AllReduceAlgo : uint32_t
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////
 
+enum class MatrixLayout
+{
+    // K-major layout (default). [Mn, K]
+    MajorK = 0,
+    // M-major for A and N-major for B. [K, Mn]
+    MajorMn,
+    // Layout is blocked along the K dimension as seen in the diagram below. [K / blockK, Mn, blockK]
+    // where blockK is fixed at 128B
+    //
+    //         ├──────────────  K  ──────────────┤
+    //  ┬  ┬   ├──── K block ───┤
+    //  │  │   │ 0   1   2   3  ║ 32  33  34  35 │
+    //  │ CTA0 │ 4   5   6   7  ║ 36  37  38  39 │
+    //  │  │   │ 8   9   10  11 ║ 40  41  42  43 │
+    //  │  ┴   │ 12  13  14  15 ║ 44  45  46  47 │
+    //  M  ┬   ├────────────────║────────────────┤
+    //  │  │   │ 16  17  18  19 ║ 48  49  50  51 │
+    //  │ CTA1 │ 20  21  22  23 ║ 52  53  54  55 │
+    //  │  │   │ 24  25  26  27 ║ 56  57  58  59 │
+    //  ┴  ┴   │ 28  29  30  31 ║ 60  61  62  63 │
+    BlockMajorK
+};
+
+////////////////////////////////////////////////////////////////////////////////////////////////////
+
 enum class SplitK : uint32_t
 {
     // No split-k is needed. I.e. mNumSlicesForSplitK == 1.
@@ -54,6 +79,20 @@ enum class SplitK : uint32_t
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////
 
+enum class BiasType : uint32_t
+{
+    // No bias.
+    None = 0,
+    // One bias value per N of the output tensor.
+    M = 1,
+    // One bias value per row M of the output tensor.
+    N = 2,
+    // One bias value for each element of the output tensor.
+    Mn = 3,
+};
+
+////////////////////////////////////////////////////////////////////////////////////////////////////
+
 enum class TileScheduler
 {
     // Static scheduler (Non-persistent).
@@ -80,6 +119,23 @@ SPLIT_K_FUNCTION(Dsmem)
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////
 
+// Helper functions to check the Bias type.
+
+#define BIAS_TYPE_FUNCTION(Mode)                                                                                       \
+    inline bool isBiasType##Mode(BiasType type)                                                                        \
+    {                                                                                                                  \
+        return (type == BiasType::Mode);                                                                               \
+    }
+
+BIAS_TYPE_FUNCTION(None)
+BIAS_TYPE_FUNCTION(N)
+BIAS_TYPE_FUNCTION(M)
+BIAS_TYPE_FUNCTION(Mn)
+
+#undef BIAS_TYPE_FUNCTION
+
+////////////////////////////////////////////////////////////////////////////////////////////////////
+
 } // namespace gemm
 
 } // namespace batchedGemm
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/GemmGatedActOptions.h b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/GemmGatedActOptions.h
index 890cb5fc54..aef650139b 100644
--- a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/GemmGatedActOptions.h
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/GemmGatedActOptions.h
@@ -59,7 +59,17 @@ namespace tg = trtllm::gen;
 enum class ActType
 {
     // silu(x) = x * sigmoid(x) = x * (1 / (1 + e^(-x)))
-    Silu = 0
+    // For ActType == Silu,
+    //    gatedAct = scaleC * x0 * silu(x1 * scaleGate),
+    // where x0 and x1 are the raw numbers from Gemm, while scaleC and scaleGate are input scales.
+    Silu = 0,
+    // For ActType == SwiGlu, ideally we would like to have something like
+    //    gatedAct = scaleC * (x0 * scaleAb + beta) * sigmoid(alpha * x1 * scaleGate).
+    // But for now, we use the simplified version
+    //    gatedAct = scaleC' * (x0 + beta') * sigmoid(alpha * x1 * scaleGate),
+    // where x0 and x1 are the raw numbers from Gemm, while scaleC and scaleGate are input scales,
+    // beta' = beta / scaleAb, scaleC' = scaleC * scaleAb.
+    SwiGlu
 };
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////
@@ -73,11 +83,24 @@ enum class ActType
     }
 
 TLLM_ACT_TYPE_FUNCTION(Silu)
+TLLM_ACT_TYPE_FUNCTION(SwiGlu)
 
 #undef TLLM_ACT_TYPE_FUNCTION
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////
 
+inline std::string getActTypeName(ActType type)
+{
+    switch (type)
+    {
+    case ActType::Silu: return "Silu";
+    case ActType::SwiGlu: return "SwiGlu";
+    default: return "Unknown type";
+    }
+}
+
+////////////////////////////////////////////////////////////////////////////////////////////////////
+
 struct GemmGatedActOptions : public gemm::GemmOptions
 {
     GemmGatedActOptions() = default;
@@ -141,6 +164,11 @@ inline bool checkAndUpdateGemmGatedActOptions(
         TLLM_CHECK_ERROR(doesSplitKUseDsmem(options.mSplitK), "Split-k GMEM and GemmGatedAct are not supported yet.");
     }
 
+    if (gemm::isBiasTypeMn(options.mBiasType))
+    {
+        TLLM_CHECK_ERROR(options.mTransposeMmaOutput, "Bias type Mn is not supported with not transpose mma output.");
+    }
+
     return true;
 }
 
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/GemmOptions.h b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/GemmOptions.h
index feb5e71969..32678fb6d3 100644
--- a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/GemmOptions.h
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/GemmOptions.h
@@ -91,20 +91,23 @@ struct GemmOptions
 
     GemmOptions() = default;
 
-    GemmOptions(AllReduceAlgo allReduceAlgo, int clusterDimX, int clusterDimY, int clusterDimZ, tg::Dtype dtypeAcc,
-        tg::Dtype dtypeA, tg::Dtype dtypeB, tg::Dtype dtypeC, bool enablesEarlyExit, bool enablesDelayedEarlyExit,
-        bool enablesGlobalPtxKnobs, int epilogueLdtmDps, int epilogueLdtmBits, int epilogueTileM, int epilogueTileN,
-        bool gridTriggerSecondaryA, bool gridTriggerSecondaryB, bool gridWaitForPrimaryEarlyExit,
-        bool gridWaitForPrimaryA, bool gridWaitForPrimaryB, bool hoistLoadTaskInit, bool hoistMmaTaskTryWaits, int k,
-        KernelTraits kernelTraits, int m, int mmaK, tg::MmaKind mmaKind, int mmaM, int mmaN, bool mockAllReduce, int n,
-        int numSlicesForSplitK, int numSlicesForSliceK, int numStages, int numStagesMma, int numStagesMmaWithinWorkTile,
-        int numStagesMmaAcrossWorkTile, int numStagesWorkId, bool outputDebugTensors, bool useShuffledMatrixA,
-        bool sliceK, SplitK splitK, bool transposeMatrixA, bool transposeMatrixB, bool transposeMmaOutput, int tileM,
-        int tileN, int tileK, bool useUnrollLoop2xForMma, bool useCustomMmaSchedule,
-        bool useHoistTryWaitForCustomMmaSchedule, bool useDeepSeekFp8, bool usePerTokenSfA, bool usePerTokenSfB,
-        bool useTmaStore, bool useTwoTmaLoadWarps, bool useTwoMmaWarps, tg::SfLayout sfLayoutA, tg::SfLayout sfLayoutB,
-        tg::SfLayout sfLayoutC, TileScheduler tileScheduler)
+    GemmOptions(AllReduceAlgo allReduceAlgo, BiasType biasType, int blockK, int clusterDimX, int clusterDimY,
+        int clusterDimZ, tg::Dtype dtypeAcc, tg::Dtype dtypeA, tg::Dtype dtypeB, tg::Dtype dtypeC, tg::Dtype dtypeMmaA,
+        tg::Dtype dtypeMmaB, bool enablesEarlyExit, bool enablesDelayedEarlyExit, bool enablesGlobalPtxKnobs,
+        int epilogueLdtmDps, int epilogueLdtmBits, int epilogueTileM, int epilogueTileN, bool gridTriggerSecondaryA,
+        bool gridTriggerSecondaryB, bool gridWaitForPrimaryEarlyExit, bool gridWaitForPrimaryA,
+        bool gridWaitForPrimaryB, bool hoistLoadTaskInit, bool hoistMmaTaskTryWaits, int k, KernelTraits kernelTraits,
+        MatrixLayout layoutA, MatrixLayout layoutB, int m, int mmaK, tg::MmaKind mmaKind, int mmaM, int mmaN,
+        bool mockAllReduce, int n, int numSlicesForSplitK, int numSlicesForSliceK, int numStages, int numStagesMma,
+        int numStagesMmaWithinWorkTile, int numStagesMmaAcrossWorkTile, int numStagesWorkId, bool outputDebugTensors,
+        bool useShuffledMatrixA, bool sliceK, SplitK splitK, bool transposeMmaOutput, int tileM, int tileN, int tileK,
+        bool useUnrollLoop2xForMma, bool useCustomMmaSchedule, bool useHoistTryWaitForCustomMmaSchedule,
+        bool useDeepSeekFp8, bool usePerTokenSfA, bool usePerTokenSfB, bool useTmaStore, bool useTwoTmaLoadWarps,
+        bool useTwoMmaWarps, tg::SfLayout sfLayoutA, tg::SfLayout sfLayoutB, tg::SfLayout sfLayoutC,
+        int sfReshapeFactor, TileScheduler tileScheduler)
         : mAllReduceAlgo{allReduceAlgo}
+        , mBiasType{biasType}
+        , mBlockK(blockK)
         , mClusterDimX{clusterDimX}
         , mClusterDimY{clusterDimY}
         , mClusterDimZ{clusterDimZ}
@@ -112,6 +115,8 @@ struct GemmOptions
         , mDtypeA{dtypeA}
         , mDtypeB{dtypeB}
         , mDtypeC{dtypeC}
+        , mDtypeMmaA{dtypeMmaA}
+        , mDtypeMmaB{dtypeMmaB}
         , mEnablesEarlyExit{enablesEarlyExit}
         , mEnablesDelayedEarlyExit{enablesDelayedEarlyExit}
         , mEnablesGlobalPtxKnobs{enablesGlobalPtxKnobs}
@@ -128,6 +133,8 @@ struct GemmOptions
         , mHoistMmaTaskTryWaits{hoistMmaTaskTryWaits}
         , mK{k}
         , mKernelTraits{kernelTraits}
+        , mLayoutA{layoutA}
+        , mLayoutB{layoutB}
         , mM{m}
         , mMmaK{mmaK}
         , mMmaKind{mmaKind}
@@ -146,8 +153,6 @@ struct GemmOptions
         , mUseShuffledMatrixA{useShuffledMatrixA}
         , mSliceK{sliceK}
         , mSplitK{splitK}
-        , mTransposeMatrixA{transposeMatrixA}
-        , mTransposeMatrixB{transposeMatrixB}
         , mTransposeMmaOutput{transposeMmaOutput}
         , mTileM{tileM}
         , mTileN{tileN}
@@ -164,13 +169,17 @@ struct GemmOptions
         , mSfLayoutA{sfLayoutA}
         , mSfLayoutB{sfLayoutB}
         , mSfLayoutC{sfLayoutC}
+        , mSfReshapeFactor{sfReshapeFactor}
         , mTileScheduler{tileScheduler}
     {
     }
 
     // The all-reduce algorithm.
     AllReduceAlgo mAllReduceAlgo{AllReduceAlgo::None};
-
+    // The type of bias.
+    BiasType mBiasType{BiasType::None};
+    // Block size in the K dimension
+    int mBlockK{-1};
     // Cluster size in X dim.
     int mClusterDimX{1};
     // Cluster size in Y dim.
@@ -185,6 +194,10 @@ struct GemmOptions
     tg::Dtype mDtypeB{tg::Dtype::Void};
     // Data type of the outputs.
     tg::Dtype mDtypeC{tg::Dtype::Void};
+    // Data type of the A matrix for the MMA, if different from the input type.
+    tg::Dtype mDtypeMmaA{tg::Dtype::Void};
+    // Data type of the B matrix for the MMA, if different from the input type.
+    tg::Dtype mDtypeMmaB{tg::Dtype::Void};
     // Whether to enable early exit.
     bool mEnablesEarlyExit{false};
     // Whether to enable delayed early exit to overlap
@@ -225,6 +238,10 @@ struct GemmOptions
     int mK{16 * 16};
     // Traits of the kernel.
     KernelTraits mKernelTraits{};
+    // Layout of A matrix
+    MatrixLayout mLayoutA{MatrixLayout::MajorK};
+    // Layout of B matrix
+    MatrixLayout mLayoutB{MatrixLayout::MajorK};
     // The M dimension of GEMM.
     int mM{128 * 2};
     // Size of the MMA instruction in the K dimension.
@@ -265,10 +282,6 @@ struct GemmOptions
     bool mSliceK{false};
     // The location of the exchange for split-K (it's None when split-K is disabled).
     SplitK mSplitK{SplitK::None};
-    // Is A matrix in a transposed layout? M major if true, K major otherwise
-    bool mTransposeMatrixA{false};
-    // Is B matrix in a transposed layout? K major if true, N major otherwise
-    bool mTransposeMatrixB{true};
     // Save output of MMA in M-major format.
     bool mTransposeMmaOutput{false};
     // M tile dimension of GEMM.
@@ -303,6 +316,12 @@ struct GemmOptions
     tg::SfLayout mSfLayoutB{tg::SfLayout::R128c4};
     // Scale factors layout for C.
     tg::SfLayout mSfLayoutC{tg::SfLayout::R128c4};
+    // Number of "repeats", i.e. reshaping factor, to fold hidden dimension into SfBlock dimension.
+    // As result, the hidden dimension of the SF tensor must be a multiple of NumRepeats *
+    // numEltsPerSf * 4. This reduces the problem shape space that the kernel is able to run.
+    // But it reduces the number of L2 requests under the hood and potentially improves perf.
+    // Applies to layout 8x4 only.
+    int mSfReshapeFactor{1};
     // Tile scheduler type.
     TileScheduler mTileScheduler{TileScheduler::Static};
 };
@@ -373,6 +392,10 @@ inline std::string dumpOptions(GemmOptions const& options)
     ss << "mAllReduceAlgo="
        << "gemm::AllReduceAlgo(" << static_cast<int32_t>(options.mAllReduceAlgo) << ")"
        << "," << std::endl;
+    ss << "mBiasType="
+       << "gemm::BiasType(" << static_cast<int32_t>(options.mBiasType) << ")"
+       << "," << std::endl;
+    ss << "mBlockK=" << options.mBlockK << "," << std::endl;
     ss << "mClusterDimX=" << options.mClusterDimX << "," << std::endl;
     ss << "mClusterDimY=" << options.mClusterDimY << "," << std::endl;
     ss << "mClusterDimZ=" << options.mClusterDimZ << "," << std::endl;
@@ -388,6 +411,12 @@ inline std::string dumpOptions(GemmOptions const& options)
     ss << "mDtypeC="
        << "trtllm::gen::Dtype(" << static_cast<int32_t>(options.mDtypeC) << ")"
        << "," << std::endl;
+    ss << "mDtypeMmaA="
+       << "trtllm::gen::Dtype(" << static_cast<int32_t>(options.mDtypeMmaA) << ")"
+       << "," << std::endl;
+    ss << "mDtypeMmaB="
+       << "trtllm::gen::Dtype(" << static_cast<int32_t>(options.mDtypeMmaB) << ")"
+       << "," << std::endl;
     ss << "mEnablesEarlyExit=" << options.mEnablesEarlyExit << "," << std::endl;
     ss << "mEnablesDelayedEarlyExit=" << options.mEnablesDelayedEarlyExit << "," << std::endl;
     ss << "mEnablesGlobalPtxKnobs=" << options.mEnablesGlobalPtxKnobs << "," << std::endl;
@@ -405,6 +434,10 @@ inline std::string dumpOptions(GemmOptions const& options)
     ss << "mK=" << options.mK << "," << std::endl;
     ss << "mKernelTraits={}"
        << "," << std::endl;
+    ss << "mLayoutA=gemm::MatrixLayout(" << static_cast<int32_t>(options.mLayoutA) << ")"
+       << "," << std::endl;
+    ss << "mLayoutB=gemm::MatrixLayout(" << static_cast<int32_t>(options.mLayoutB) << ")"
+       << "," << std::endl;
     ss << "mM=" << options.mM << "," << std::endl;
     ss << "mMmaK=" << options.mMmaK << "," << std::endl;
     ss << "mMmaKind="
@@ -427,8 +460,6 @@ inline std::string dumpOptions(GemmOptions const& options)
     ss << "mSplitK="
        << "gemm::SplitK(" << static_cast<int32_t>(options.mSplitK) << ")"
        << "," << std::endl;
-    ss << "mTransposeMatrixA=" << options.mTransposeMatrixA << "," << std::endl;
-    ss << "mTransposeMatrixB=" << options.mTransposeMatrixB << "," << std::endl;
     ss << "mTransposeMmaOutput=" << options.mTransposeMmaOutput << "," << std::endl;
     ss << "mTileM=" << options.mTileM << "," << std::endl;
     ss << "mTileN=" << options.mTileN << "," << std::endl;
@@ -451,6 +482,7 @@ inline std::string dumpOptions(GemmOptions const& options)
     ss << "mSfLayoutC="
        << "trtllm::gen::SfLayout(" << static_cast<int32_t>(options.mSfLayoutC) << ")"
        << "," << std::endl;
+    ss << "mSfReshapeFactor=" << options.mSfReshapeFactor << "," << std::endl;
     ss << "mTileScheduler="
        << "gemm::TileScheduler(" << static_cast<int32_t>(options.mTileScheduler) << ")" << std::endl;
     return ss.str();
@@ -502,39 +534,89 @@ inline bool checkAndUpdateGemmOptions(
         }
     }
 
+    // If not specified, used the input dtypes as MMA dtypes (no cast required).
+    if (options.mDtypeMmaA == tg::Dtype::Void)
+    {
+        if (updateOptions)
+        {
+            options.mDtypeMmaA = options.mDtypeA;
+        }
+        else
+        {
+            return false;
+        }
+    }
+    if (options.mDtypeMmaB == tg::Dtype::Void)
+    {
+        if (updateOptions)
+        {
+            options.mDtypeMmaB = options.mDtypeB;
+        }
+        else
+        {
+            return false;
+        }
+    }
+
+    // Check that the A cast is supported.
+    // Currently, we only support {MxFp4, NvFp4} -> Bf16.
+    TLLM_CHECK_ERROR((options.mDtypeA == options.mDtypeMmaA)
+            || ((options.mDtypeA == tg::Dtype::MxE2m1 || options.mDtypeA == tg::Dtype::E2m1)
+                && options.mDtypeMmaA == tg::Dtype::Bfloat16),
+        "Unsupported cast: ", tg::dtypeToString(options.mDtypeA), " -> ", tg::dtypeToString(options.mDtypeMmaA));
+    // Casting B is currently not supported.
+    // Note: This is because we currently write the output of the cast directly to TMEM.
+    //       In the future, we can support this through shared memory.
+    TLLM_CHECK_ERROR(
+        (options.mDtypeB == options.mDtypeMmaB), "Casting B is not supported: dtypeMmaB must be the same as dtypeB.");
+
+    if (options.mDtypeA != options.mDtypeMmaA)
+    {
+        TLLM_CHECK_ERROR(options.mTileM == 128, "TileM must be 128 when casting the input matrix A before the MMA.");
+    }
+
     // FIXME: We do not support different dtypes for A and B when not on Blackwell.
     if (!isBlackwell)
     {
-        TLLM_CHECK_ERROR(options.mDtypeA == options.mDtypeB, "For non-Blackwell, A and B must have the same dtype.");
+        TLLM_CHECK_ERROR(
+            options.mDtypeMmaA == options.mDtypeMmaB, "For non-Blackwell, A and B must have the same dtype.");
     }
 
     // Check that the different dtypes for A and B are supported by the tensor core
     // kind::f8f6f4
-    if (options.mDtypeA == tg::Dtype::E4m3 || options.mDtypeA == tg::Dtype::E2m1)
+    if (options.mDtypeMmaA == tg::Dtype::E4m3 || options.mDtypeMmaA == tg::Dtype::E2m1)
     {
-        TLLM_CHECK_ERROR(options.mDtypeB == tg::Dtype::E4m3 || options.mDtypeB == tg::Dtype::E2m1,
-            "For E4m3/E2m1 A, B must also be E4m3/E2m1.");
+        TLLM_CHECK_ERROR(options.mDtypeMmaB == tg::Dtype::E4m3 || options.mDtypeMmaB == tg::Dtype::E2m1,
+            "For dtypeMmaA = E4m3/E2m1 A, dtypeMmaB must also be E4m3/E2m1.");
     }
 
     // kind::mxf8f6f4
-    if (options.mDtypeA == tg::Dtype::MxE4m3 || options.mDtypeA == tg::Dtype::MxE2m1)
+    if (options.mDtypeMmaA == tg::Dtype::MxE4m3 || options.mDtypeMmaA == tg::Dtype::MxE2m1)
     {
-        TLLM_CHECK_ERROR(options.mDtypeB == tg::Dtype::MxE4m3 || options.mDtypeB == tg::Dtype::MxE2m1,
-            "For dtypeA = MxE4m3 or MxE2m1, dtypeB must also be MxE4m3 or MxE2m1.");
+        TLLM_CHECK_ERROR(options.mDtypeMmaB == tg::Dtype::MxE4m3 || options.mDtypeMmaB == tg::Dtype::MxE2m1,
+            "For dtypeMmaA = MxE4m3 or MxE2m1, dtypeMmaB must also be MxE4m3 or MxE2m1.");
     }
-    if (options.mDtypeB == tg::Dtype::MxE4m3 || options.mDtypeB == tg::Dtype::MxE2m1)
+    if (options.mDtypeMmaB == tg::Dtype::MxE4m3 || options.mDtypeMmaB == tg::Dtype::MxE2m1)
     {
-        TLLM_CHECK_ERROR(options.mDtypeA == tg::Dtype::MxE4m3 || options.mDtypeA == tg::Dtype::MxE2m1,
-            "For dtypeB = MxE4m3 or MxE2m1, dtypeA must also be MxE4m3 or MxE2m1.");
+        TLLM_CHECK_ERROR(options.mDtypeMmaA == tg::Dtype::MxE4m3 || options.mDtypeMmaA == tg::Dtype::MxE2m1,
+            "For dtypeMmaB = MxE4m3 or MxE2m1, dtypeMmaA must also be MxE4m3 or MxE2m1.");
     }
 
     // kind::f16
-    if (options.mDtypeA == tg::Dtype::Fp16 || options.mDtypeA == tg::Dtype::Bfloat16)
+    if (options.mDtypeMmaA == tg::Dtype::Fp16 || options.mDtypeMmaA == tg::Dtype::Bfloat16)
     {
-        TLLM_CHECK_ERROR(options.mDtypeB == options.mDtypeA, "For Fp16/Bfloat16 A, B must be the same type as A.");
+        TLLM_CHECK_ERROR(options.mDtypeMmaB == options.mDtypeMmaA,
+            "For dtypeMmaA = Fp16/Bfloat16, dtypeMmaB must be the same as dtypeMmaA.");
     }
 
-    // When different dtype are used for A and B, we must use different tile to do the loading.
+    // When one of the inputs needs to be cast, we must use two load warps.
+    if ((options.mDtypeMmaA != options.mDtypeA || options.mDtypeMmaB != options.mDtypeB)
+        && !options.mUseTwoTmaLoadWarps)
+    {
+        TLLM_LOG_WARNING("Two TMA load warps must be enabled if any of the inputs needs to be cast.");
+    }
+
+    // When different dtypes are used for A and B, we must use different tiles to do the loading.
     // It is not strictly required, but current implementation of SmemAb requires that.
     if (options.mDtypeA != options.mDtypeB)
     {
@@ -547,7 +629,7 @@ inline bool checkAndUpdateGemmOptions(
     {
         if (updateOptions)
         {
-            options.mMmaKind = dtypeGetMmaKind(options.mDtypeA, options.mDtypeB);
+            options.mMmaKind = dtypeGetMmaKind(options.mDtypeMmaA, options.mDtypeMmaB);
         }
         else
         {
@@ -555,11 +637,6 @@ inline bool checkAndUpdateGemmOptions(
         }
     }
 
-    if (options.mMmaKind == tg::MmaKind::Fp16)
-    {
-        TLLM_CHECK_ERROR(options.mDtypeA == options.mDtypeB, "For Fp16 MMA, A and B must have the same dtype.");
-    }
-
     if ((options.mMmaKind == tg::MmaKind::Fp8Fp6Fp4 || options.mMmaKind == tg::MmaKind::MxFp8Fp6Fp4)
         && options.mMmaK != 32)
     {
@@ -626,7 +703,8 @@ inline bool checkAndUpdateGemmOptions(
     {
         TLLM_CHECK_ERROR(isBlackwell, "Block scaling is only supported on Blackwell");
 
-        TLLM_CHECK_ERROR(options.mSfLayoutB == tg::SfLayout::R128c4 || options.mSfLayoutB == tg::SfLayout::R8c4,
+        TLLM_CHECK_ERROR(options.mSfLayoutB == tg::SfLayout::R128c4 || options.mSfLayoutB == tg::SfLayout::R8c4
+                || options.mSfLayoutB == tg::SfLayout::Linear,
             "Only the 128x4 and 8x4 SF layouts are supported for B, got ", tg::sfLayoutToString(options.mSfLayoutB));
 
         int const mmaK = (options.mMmaKind == tg::MmaKind::MxFp4NvFp4) ? 64 : 32;
@@ -653,11 +731,16 @@ inline bool checkAndUpdateGemmOptions(
         // The MMA N may only be smaller than 64 if it is equal to the tile N.
         TLLM_CHECK_ERROR(options.mMmaN >= 64 || options.mMmaN == options.mTileN, "MmaN (", options.mMmaN,
             ") must be >= 64 or equal to TileN (", options.mTileN, ")");
-
+    }
+    if (tg::dtypeIsBlockFmt(options.mDtypeA))
+    {
         int numEltsPerSfA = tg::dtypeNumEltsPerSf(options.mDtypeA);
-        int numEltsPerSfB = tg::dtypeNumEltsPerSf(options.mDtypeB);
         TLLM_CHECK_ERROR(options.mTileK % (4 * numEltsPerSfA) == 0, "TileK (", options.mTileK,
             ") must be a multiple of ", (4 * numEltsPerSfA), " for typeA ", gemm::toString(options.mDtypeA));
+    }
+    if (tg::dtypeIsBlockFmt(options.mDtypeB))
+    {
+        int numEltsPerSfB = tg::dtypeNumEltsPerSf(options.mDtypeB);
         TLLM_CHECK_ERROR(options.mTileK % (4 * numEltsPerSfB) == 0, "TileK (", options.mTileK,
             ") must be a multiple of ", (4 * numEltsPerSfB), " for typeB ", gemm::toString(options.mDtypeB));
     }
@@ -753,7 +836,6 @@ inline bool checkAndUpdateGemmOptions(
 
     TLLM_CHECK_ERROR(options.mM > 0 && options.mN > 0 && options.mK > 0, "M, N and K must be larger than 0");
     TLLM_CHECK_ERROR(options.mNumSlicesForSplitK > 0, "Split K must be larger than 0.");
-    TLLM_CHECK_ERROR(options.mK % options.mTileK == 0, "K must be a multiple of TileK");
 
     if (options.mUseShuffledMatrixA)
     {
@@ -911,6 +993,11 @@ inline bool checkAndUpdateGemmOptions(
     {
         TLLM_CHECK_ERROR(
             options.mNumStagesMmaWithinWorkTile == 1, "Non-DeepSeekFp8 requires numStagesMmaWithinWorkTile == 1");
+        if (options.mNumStagesMma > 1)
+        {
+            TLLM_CHECK_ERROR(options.mTileScheduler == TileScheduler::Persistent,
+                "Non-DeepSeekFp8 requires persistent scheduler when using numStagesMma >1");
+        }
     }
     if (options.mUseDeepSeekFp8)
     {
@@ -923,6 +1010,7 @@ inline bool checkAndUpdateGemmOptions(
 
         // Check that TileK = 128 for correct scaling of every 128 channels.
         TLLM_CHECK_ERROR(options.mTileK == 128, "Tile-K must be equal to 128 for DeepSeek Fp8");
+        TLLM_CHECK_ERROR(options.mK % options.mTileK == 0, "K must be a multiple of TileK");
         // Tile sizes of the output hidden dimension.
         auto hiddenDimPerOutputTile = options.mTransposeMmaOutput ? options.mTileM : options.mTileN;
         auto hiddenDimPerEpilogueTile = options.mTransposeMmaOutput ? options.mEpilogueTileM : options.mEpilogueTileN;
@@ -997,14 +1085,22 @@ inline bool checkAndUpdateGemmOptions(
 
     if (options.mUseUnrollLoop2xForMma)
     {
-        bool notSupported = (options.mK / options.mNumSlicesForSplitK) % (options.mTileK * 2) != 0;
-        // Check that the 2*TileK is a multiple of MmaK when UnrollLoop2x is enabled.
-        // This is to avoid deadlock when mma runs even-numbered loop while the other warps run
-        // odd-numbered loop.
+        // Number of iterations in K dimension after padding.
+        // Note the perCtaK in each CTA in the splitK group are padded to the same number of iterations.
+        // E.g., K = 512, TileK = 128, numSlicesForSplitK = 3. Then the padded K is
+        //
+        //   ceil(512 / (128*3)) * (128*3) = 768
+        //
+        int paddedK = divUpMul(options.mK, options.mTileK * options.mNumSlicesForSplitK);
+        // Check that the padded K (K rounded to next multiple of tileK) is a multiple of 2*TileK when
+        // UnrollLoop2x is enabled. This is to avoid deadlock when mma runs even-numbered loop while the
+        // other warps run odd-numbered loop.
+        //
+        bool notSupported = (paddedK / options.mNumSlicesForSplitK) % (options.mTileK * 2) != 0;
         if (notSupported)
         {
             TLLM_LOG_WARNING("Size K / splitK must be a multiple of TileK * 2. Found TileK=", options.mTileK,
-                " and K=", options.mK, " and numSlicesForSplitK=", options.mNumSlicesForSplitK,
+                " and K=", options.mK, " (paddedK=", paddedK, ") and numSlicesForSplitK=", options.mNumSlicesForSplitK,
                 ". Disabling unrollLoop2xForMma.");
             if (updateOptions)
             {
@@ -1059,43 +1155,108 @@ inline bool checkAndUpdateGemmOptions(
     //
     // Kernel 1:  ----PREEXIT-----------FLUSH
     // Kernel 2:      -------PREEXIT----ACQBULK---FLUSH
-    // Kernel 3:  Warp 0:           ---- (!) Output of 1,2 is not yet visible -----------------------
-    //            Warp 1:           ---- (!) We normally assume that 1 is visible is not yet visible-
-    //            Warp 2:           -------------------ACQBULK-- Kernel 1,2 output visible ----------
+    // Kernel 3:  Warp 0:           ---- (!) Output of 1,2 is not yet visible
+    // -----------------------
+    //            Warp 1:           ---- (!) We normally assume that 1 is visible is not yet
+    //            visible- Warp 2:           -------------------ACQBULK-- Kernel 1,2 output visible
+    //            ----------
     TLLM_CHECK_ERROR((options.mGridWaitForPrimaryA || !options.mGridTriggerSecondaryA),
         "A: If a task triggers a secondary kernel, it must also wait for primary kernel.");
     TLLM_CHECK_ERROR((options.mGridWaitForPrimaryB || !options.mGridTriggerSecondaryB),
         "B: If a task triggers a secondary kernel, it must also wait for primary kernel.");
 
+    if (options.mUsePerTokenSfA || options.mUsePerTokenSfB)
+    {
+        // Checks applicable to both MetaFP8 and RoutingScalesOnInput
+        TLLM_CHECK_ERROR(!options.mUseDeepSeekFp8, "DeepSeek FP8 and per-token scaling are not compatible");
+        TLLM_CHECK_ERROR(isBlackwell, "Per-token scaling is not supported for Hopper");
+        if (options.mUsePerTokenSfA && options.mUsePerTokenSfB)
+        {
+            // MetaFP8 case
+            TLLM_CHECK_ERROR(options.mDtypeA == tg::Dtype::E4m3 && options.mDtypeB == tg::Dtype::E4m3,
+                "A and B dtype must be E4m3 for Meta Fp8. Found dtypeA=", tg::dtypeToString(options.mDtypeA),
+                " dtypeB=", tg::dtypeToString(options.mDtypeB));
+        }
+        else
+        {
+            // RoutingScalesOnInput case
+            TLLM_CHECK_ERROR((options.mUsePerTokenSfA && !options.mTransposeMmaOutput)
+                    || (options.mUsePerTokenSfB && options.mTransposeMmaOutput),
+                "In RoutingScalesOnInput mode, perToken scales must be used on activations");
+        }
+    }
+
     // The generation should support non K-major layouts for both A and B; however, it is unclear if
     // there is a use-case
-    TLLM_CHECK_ERROR(!options.mTransposeMatrixA || options.mTransposeMatrixB,
-        "TransposeA true and TransposeB false is not supported");
+    TLLM_CHECK_ERROR((options.mLayoutA == MatrixLayout::MajorK) || (options.mLayoutB == MatrixLayout::MajorK),
+        "At least one matrix must be in k-major layout");
 
     // Some features are currently only support when both matrices are in K-major format
-    if (options.mTransposeMatrixA || !options.mTransposeMatrixB)
+    if (options.mLayoutB != MatrixLayout::MajorK || options.mLayoutB != MatrixLayout::MajorK)
     {
         TLLM_CHECK_ERROR(isBlackwell, "Non K-major layouts are only supported on Blackwell");
         TLLM_CHECK_ERROR(options.mSplitK == SplitK::None, "Non K-major layouts do not support split K");
     }
-    if (options.mTransposeMatrixA)
+    if (options.mLayoutA == MatrixLayout::MajorMn)
     {
         TLLM_CHECK_ERROR(tg::dtypeGetNumBits(options.mDtypeA) >= 8, "Subbyte types only support K major layout");
     }
-    if (!options.mTransposeMatrixB)
+    if (options.mLayoutB == MatrixLayout::MajorMn)
     {
         TLLM_CHECK_ERROR(tg::dtypeGetNumBits(options.mDtypeB) >= 8, "Subbyte types only support K major layout");
     }
 
+    if ((options.mLayoutA == MatrixLayout::BlockMajorK) || (options.mLayoutB == MatrixLayout::BlockMajorK))
+    {
+        bool const isBlockA = options.mLayoutA == MatrixLayout::BlockMajorK;
+
+        // Block K size must be 128B.
+        // TODO Leaving this as an option for now in case we want to expertiment with other block sizes
+        // As the user is not expected to set this, do not fail if updateOptions is false
+        int32_t const elemSizeInBits
+            = (isBlockA) ? tg::dtypeGetNumBits(options.mDtypeA) : tg::dtypeGetNumBits(options.mDtypeB);
+        int32_t const elemsIn128B = 128 * 8 /* Bits in byte */ / elemSizeInBits;
+
+        if (options.mBlockK != elemsIn128B)
+        {
+            if (updateOptions)
+            {
+                options.mBlockK = elemsIn128B;
+            }
+            else
+            {
+                return false;
+            }
+        }
+
+        if (options.mBlockK > options.mTileK)
+        {
+            TLLM_CHECK_ERROR(options.mBlockK % options.mTileK == 0,
+                "If block size is greater than tile size, block size must be a multiple of tile size");
+        }
+        else if (options.mBlockK < options.mTileK)
+        {
+            TLLM_CHECK_ERROR(options.mTileK % options.mBlockK == 0,
+                "If tile size is greater than block size, tile size must be a multiple of block size");
+        }
+    }
+
+    if (!isBiasTypeNone(options.mBiasType))
+    {
+        TLLM_CHECK_ERROR(!isBiasTypeMn(options.mBiasType), "BiasType::Mn is not supported");
+        TLLM_CHECK_ERROR(!options.mUseDeepSeekFp8, "Bias is not supported for DeepSeek Fp8");
+        TLLM_CHECK_ERROR(!(options.mUsePerTokenSfA && options.mUsePerTokenSfB), "Bias is not supported for Meta Fp8");
+    }
+
     if (updateOptions)
     {
         // Init kernel traits.
         options.mKernelTraits = KernelTraits(options.mDtypeA, options.mDtypeB, options.mDtypeC, options.mDtypeAcc,
-            options.mMmaKind, options.mTileM, options.mTileN, options.mTileK, options.mEpilogueTileM,
-            options.mEpilogueTileN, options.mNumStages, options.mNumStagesMma, options.mNumSlicesForSplitK,
-            options.mNumSlicesForSliceK, options.mSplitK, options.mUseTmaStore, options.mTransposeMmaOutput,
-            options.mAllReduceAlgo, options.mTileScheduler == TileScheduler::Persistent, options.mUseDeepSeekFp8,
-            options.mUsePerTokenSfA, options.mUsePerTokenSfB);
+            options.mDtypeMmaA, options.mDtypeMmaB, options.mMmaKind, options.mTileM, options.mTileN, options.mTileK,
+            options.mEpilogueTileM, options.mEpilogueTileN, options.mNumStages, options.mNumStagesMma,
+            options.mNumSlicesForSplitK, options.mNumSlicesForSliceK, options.mSplitK, options.mUseTmaStore,
+            options.mTransposeMmaOutput, options.mAllReduceAlgo, options.mTileScheduler == TileScheduler::Persistent,
+            options.mUseDeepSeekFp8, options.mUsePerTokenSfA, options.mUsePerTokenSfB, options.mBiasType);
     }
 
     return true;
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/KernelMetaInfo.h b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/KernelMetaInfo.h
index 61b166a2f3..307f50fbc8 100644
--- a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/KernelMetaInfo.h
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/KernelMetaInfo.h
@@ -28,169 +28,187 @@ namespace kernels
 {
 // clang-format off
 
-#define TLLM_GEN_COMMIT "2b209663"
+#define TLLM_GEN_COMMIT "e3f037b7-dirty"
 #define TLLM_GEN_EXPORT_VERSION "6.0.3.0.2.1"
 
-static constexpr size_t tllmGenBatchedGemmListLen = 74;
+static constexpr size_t tllmGenBatchedGemmListLen = 82;
 
 #ifndef EXCLUDE_SM_100
-extern unsigned char BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x256_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x256_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x256u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x256u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E2m1E2m1Fp32_E2m1_Tile128x8x512_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_InplaceRoute_GatedAct_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E2m1E2m1Fp32_E2m1_Tile128x8x512_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E2m1E2m1Fp32_E2m1_Tile128x8x512u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_InplaceRoute_GatedAct_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E2m1E2m1Fp32_E2m1_Tile128x8x512u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Fp16_Tile128x8x512_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Fp16_Tile128x8x512u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x128x128_EpiTile64x128_Mma64x128x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x128x128_EpiTile64x128_Mma64x128x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x128x128u2_EpiTile64x128_Mma64x128x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x128x128u2_EpiTile64x128_Mma64x128x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x16x128_EpiTile64x16_Mma64x16x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x16x128_EpiTile64x16_Mma64x16x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x16x128u2_EpiTile64x16_Mma64x16x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x16x128u2_EpiTile64x16_Mma64x16x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x32x128_EpiTile64x32_Mma64x32x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x32x128_EpiTile64x32_Mma64x32x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x32x128u2_EpiTile64x32_Mma64x32x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x32x128u2_EpiTile64x32_Mma64x32x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x64x128_EpiTile64x64_Mma64x64x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x64x128_EpiTile64x64_Mma64x64x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x64x128u2_EpiTile64x64_Mma64x64x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x64x128u2_EpiTile64x64_Mma64x64x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x128_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x128_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x128_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x128u2_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x128u2_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x128u2_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x128x128_EpiTile64x128_Mma64x128x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x128x128u2_EpiTile64x128_Mma64x128x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x16x128_EpiTile64x16_Mma64x16x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x16x128u2_EpiTile64x16_Mma64x16x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x32x128_EpiTile64x32_Mma64x32x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x32x128u2_EpiTile64x32_Mma64x32x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x64x128_EpiTile64x64_Mma64x64x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x64x128u2_EpiTile64x64_Mma64x64x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x128_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x128_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x128u2_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x128u2_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_InplaceRoute_GatedAct_perTokenSfB_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_InplaceRoute_GatedAct_perTokenSfB_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Fp16_Tile128x8x128_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Fp16_Tile128x8x128u2_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Fp16_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Fp16_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x256_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x256_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x256u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x256u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_MxE4m3_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_InplaceRoute_GatedAct_dynamic_sm100a_cubin[];
-extern unsigned char BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_MxE4m3_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_InplaceRoute_GatedAct_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x128x128_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x128x128_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x128x128u2_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x128x128u2_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x16x128_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x16x128_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x16x128u2_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x16x128u2_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x32x128_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x32x128_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x32x128u2_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x32x128u2_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x64x128_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x64x128_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x64x128u2_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x64x128u2_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x256u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x256u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x128x128_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x128x128u2_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x16x128_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x16x128u2_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x32x128_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x32x128u2_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x64x128_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x64x128u2_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Fp16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Fp16_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Fp16_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Fp16_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Fp16_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_Fp16_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_MxE4m3_MxE2m1MxE4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin[];
+extern unsigned char BatchedGemmKernel_MxE4m3_MxE2m1MxE4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin[];
 #endif // EXCLUDE_SM_100
 
 #ifndef EXCLUDE_SM_100
-extern unsigned int BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x256_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x256_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x256u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x256u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E2m1E2m1Fp32_E2m1_Tile128x8x512_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_InplaceRoute_GatedAct_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E2m1E2m1Fp32_E2m1_Tile128x8x512_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E2m1E2m1Fp32_E2m1_Tile128x8x512u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_InplaceRoute_GatedAct_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E2m1E2m1Fp32_E2m1_Tile128x8x512u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Fp16_Tile128x8x512_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Fp16_Tile128x8x512u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x128x128_EpiTile64x128_Mma64x128x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x128x128_EpiTile64x128_Mma64x128x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x128x128u2_EpiTile64x128_Mma64x128x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x128x128u2_EpiTile64x128_Mma64x128x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x16x128_EpiTile64x16_Mma64x16x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x16x128_EpiTile64x16_Mma64x16x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x16x128u2_EpiTile64x16_Mma64x16x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x16x128u2_EpiTile64x16_Mma64x16x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x32x128_EpiTile64x32_Mma64x32x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x32x128_EpiTile64x32_Mma64x32x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x32x128u2_EpiTile64x32_Mma64x32x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x32x128u2_EpiTile64x32_Mma64x32x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x64x128_EpiTile64x64_Mma64x64x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x64x128_EpiTile64x64_Mma64x64x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x64x128u2_EpiTile64x64_Mma64x64x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x64x128u2_EpiTile64x64_Mma64x64x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x128_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x128_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x128_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x128u2_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x128u2_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x128u2_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x128x128_EpiTile64x128_Mma64x128x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x128x128u2_EpiTile64x128_Mma64x128x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x16x128_EpiTile64x16_Mma64x16x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x16x128u2_EpiTile64x16_Mma64x16x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x32x128_EpiTile64x32_Mma64x32x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x32x128u2_EpiTile64x32_Mma64x32x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x64x128_EpiTile64x64_Mma64x64x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x64x128u2_EpiTile64x64_Mma64x64x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x128_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x128_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x128u2_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x128u2_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_InplaceRoute_GatedAct_perTokenSfB_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_InplaceRoute_GatedAct_perTokenSfB_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Fp16_Tile128x8x128_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Fp16_Tile128x8x128u2_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Fp16_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Fp16_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x256_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x256_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x256u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x256u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_MxE4m3_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_InplaceRoute_GatedAct_dynamic_sm100a_cubin_len;
-extern unsigned int BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_MxE4m3_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_InplaceRoute_GatedAct_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x128x128_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x128x128_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x128x128u2_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x128x128u2_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x16x128_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x16x128_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x16x128u2_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x16x128u2_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x32x128_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x32x128_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x32x128u2_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x32x128u2_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x64x128_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x64x128_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x64x128u2_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x64x128u2_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x256u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x256u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x128x128_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x128x128u2_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x16x128_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x16x128u2_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x32x128_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x32x128u2_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x64x128_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x64x128u2_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Fp16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Fp16_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Fp16_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Fp16_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Fp16_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_Fp16_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_MxE4m3_MxE2m1MxE4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin_len;
+extern unsigned int BatchedGemmKernel_MxE4m3_MxE2m1MxE4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin_len;
 #endif // EXCLUDE_SM_100
 
 
 static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 #ifndef EXCLUDE_SM_100
-{BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x256_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x256_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin_len, 126976, "BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x256_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a", 448, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin_len, 126976, "BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a", 448, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -198,6 +216,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(17826818)
 , /* mDtypeB */ trtllm::gen::Dtype(17826818)
 , /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(17826818)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(17826818)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -214,6 +234,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 64
 , /* mMmaKind */ trtllm::gen::MmaKind(4)
@@ -224,16 +246,14 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumSlicesForSplitK */ 1
 , /* mNumSlicesForSliceK */ 1
 , /* mNumStages */ 6
-, /* mNumStagesMma */ 2
+, /* mNumStagesMma */ 1
 , /* mNumStagesMmaWithinWorkTile */ 1
-, /* mNumStagesMmaAcrossWorkTile */ 2
+, /* mNumStagesMmaAcrossWorkTile */ 1
 , /* mNumStagesWorkId */ 3
 , /* mOutputDebugTensors */ 0
 , /* mUseShuffledMatrixA */ 1
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
@@ -250,6 +270,7 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(1)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(1)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -258,13 +279,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 0
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x256_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x256_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin_len, 126976, "BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x256_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a", 480, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin_len, 126976, "BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a", 480, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -272,6 +296,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(17826818)
 , /* mDtypeB */ trtllm::gen::Dtype(17826818)
 , /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(17826818)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(17826818)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -288,6 +314,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 64
 , /* mMmaKind */ trtllm::gen::MmaKind(4)
@@ -306,8 +334,6 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseShuffledMatrixA */ 1
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
@@ -324,6 +350,7 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(1)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(1)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(1)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -332,13 +359,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 0
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x256u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x256u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin_len, 126976, "BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x256u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a", 448, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin_len, 88064, "BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a", 480, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -346,6 +376,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(17826818)
 , /* mDtypeB */ trtllm::gen::Dtype(17826818)
 , /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(17826818)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(17826818)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -362,6 +394,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 64
 , /* mMmaKind */ trtllm::gen::MmaKind(4)
@@ -371,7 +405,7 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mN */ 256
 , /* mNumSlicesForSplitK */ 1
 , /* mNumSlicesForSliceK */ 1
-, /* mNumStages */ 6
+, /* mNumStages */ 4
 , /* mNumStagesMma */ 2
 , /* mNumStagesMmaWithinWorkTile */ 1
 , /* mNumStagesMmaAcrossWorkTile */ 2
@@ -380,13 +414,11 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseShuffledMatrixA */ 1
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
 , /* mTileK */ 256
-, /* mUseUnrollLoop2xForMma */ 1
+, /* mUseUnrollLoop2xForMma */ 0
 , /* mUseCustomMmaSchedule */ 1
 , /* mUseHoistTryWaitForCustomMmaSchedule */ 0
 , /* mUseDeepSeekFp8 */ 0
@@ -398,21 +430,25 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(1)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(1)
-, /* mTileScheduler */ gemm::TileScheduler(0)
+, /* mSfReshapeFactor */ 1
+, /* mTileScheduler */ gemm::TileScheduler(1)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
 , /* mBatchedN */ {}
 , /* mBatchMode */ batchedGemm::BatchedGemmOptions::BatchMode(1)
-, /* mNumBatches */ 2
+, /* mNumBatches */ 128
 , /* mIsStaticBatch */ 0
-, /* mNumTokens */ 2
-, /* mRouteAct */ 0
+, /* mNumTokens */ 32
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 0
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x256u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x256u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin_len, 126976, "BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x256u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a", 480, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin_len, 126976, "BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a", 448, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -420,6 +456,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(17826818)
 , /* mDtypeB */ trtllm::gen::Dtype(17826818)
 , /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(17826818)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(17826818)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -436,6 +474,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 64
 , /* mMmaKind */ trtllm::gen::MmaKind(4)
@@ -446,16 +486,14 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumSlicesForSplitK */ 1
 , /* mNumSlicesForSliceK */ 1
 , /* mNumStages */ 6
-, /* mNumStagesMma */ 2
+, /* mNumStagesMma */ 1
 , /* mNumStagesMmaWithinWorkTile */ 1
-, /* mNumStagesMmaAcrossWorkTile */ 2
+, /* mNumStagesMmaAcrossWorkTile */ 1
 , /* mNumStagesWorkId */ 3
 , /* mOutputDebugTensors */ 0
 , /* mUseShuffledMatrixA */ 1
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
@@ -472,7 +510,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(1)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(1)
-, /* mTileScheduler */ gemm::TileScheduler(1)
+, /* mSfReshapeFactor */ 1
+, /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
 , /* mBatchedN */ {}
@@ -480,13 +519,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 0
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin_len, 166912, "BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a", 448, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin_len, 126976, "BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a", 480, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -494,6 +536,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(17826818)
 , /* mDtypeB */ trtllm::gen::Dtype(17826818)
 , /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(17826818)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(17826818)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -510,6 +554,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 64
 , /* mMmaKind */ trtllm::gen::MmaKind(4)
@@ -519,7 +565,7 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mN */ 256
 , /* mNumSlicesForSplitK */ 1
 , /* mNumSlicesForSliceK */ 1
-, /* mNumStages */ 4
+, /* mNumStages */ 6
 , /* mNumStagesMma */ 2
 , /* mNumStagesMmaWithinWorkTile */ 1
 , /* mNumStagesMmaAcrossWorkTile */ 2
@@ -528,13 +574,11 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseShuffledMatrixA */ 1
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
-, /* mTileK */ 512
-, /* mUseUnrollLoop2xForMma */ 0
+, /* mTileK */ 256
+, /* mUseUnrollLoop2xForMma */ 1
 , /* mUseCustomMmaSchedule */ 1
 , /* mUseHoistTryWaitForCustomMmaSchedule */ 0
 , /* mUseDeepSeekFp8 */ 0
@@ -546,7 +590,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(1)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(1)
-, /* mTileScheduler */ gemm::TileScheduler(0)
+, /* mSfReshapeFactor */ 1
+, /* mTileScheduler */ gemm::TileScheduler(1)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
 , /* mBatchedN */ {}
@@ -554,13 +599,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 0
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin_len, 166912, "BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a", 480, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin_len, 166912, "BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a", 448, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -568,6 +616,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(17826818)
 , /* mDtypeB */ trtllm::gen::Dtype(17826818)
 , /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(17826818)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(17826818)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -584,6 +634,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 64
 , /* mMmaKind */ trtllm::gen::MmaKind(4)
@@ -594,16 +646,14 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumSlicesForSplitK */ 1
 , /* mNumSlicesForSliceK */ 1
 , /* mNumStages */ 4
-, /* mNumStagesMma */ 2
+, /* mNumStagesMma */ 1
 , /* mNumStagesMmaWithinWorkTile */ 1
-, /* mNumStagesMmaAcrossWorkTile */ 2
+, /* mNumStagesMmaAcrossWorkTile */ 1
 , /* mNumStagesWorkId */ 3
 , /* mOutputDebugTensors */ 0
 , /* mUseShuffledMatrixA */ 1
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
@@ -620,7 +670,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(1)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(1)
-, /* mTileScheduler */ gemm::TileScheduler(1)
+, /* mSfReshapeFactor */ 1
+, /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
 , /* mBatchedN */ {}
@@ -628,13 +679,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 0
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_sm100a_cubin, BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_sm100a_cubin_len, 166912, "BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_sm100a", 448, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin_len, 166912, "BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a", 448, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -642,7 +696,9 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(17826818)
 , /* mDtypeB */ trtllm::gen::Dtype(17826818)
 , /* mDtypeC */ trtllm::gen::Dtype(1052672)
-, /* mEnablesEarlyExit */ 0
+, /* mDtypeMmaA */ trtllm::gen::Dtype(17826818)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(17826818)
+, /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
 , /* mEpilogueLdtmDps */ 16
@@ -658,6 +714,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 64
 , /* mMmaKind */ trtllm::gen::MmaKind(4)
@@ -676,8 +734,6 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseShuffledMatrixA */ 1
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
@@ -694,21 +750,25 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(1)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(1)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
 , /* mBatchedN */ {}
 , /* mBatchMode */ batchedGemm::BatchedGemmOptions::BatchMode(1)
-, /* mNumBatches */ 2
-, /* mIsStaticBatch */ 1
-, /* mNumTokens */ 0
-, /* mRouteAct */ 0
-, /* mGridWaitForPrimaryRouting */ 1
+, /* mNumBatches */ 128
+, /* mIsStaticBatch */ 0
+, /* mNumTokens */ 32
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
+, /* mGridWaitForPrimaryRouting */ 0
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin_len, 166912, "BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a", 448, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin, BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin_len, 166912, "BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a", 448, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -716,7 +776,9 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(17826818)
 , /* mDtypeB */ trtllm::gen::Dtype(17826818)
 , /* mDtypeC */ trtllm::gen::Dtype(1052672)
-, /* mEnablesEarlyExit */ 1
+, /* mDtypeMmaA */ trtllm::gen::Dtype(17826818)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(17826818)
+, /* mEnablesEarlyExit */ 0
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
 , /* mEpilogueLdtmDps */ 16
@@ -732,6 +794,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 64
 , /* mMmaKind */ trtllm::gen::MmaKind(4)
@@ -742,21 +806,19 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumSlicesForSplitK */ 1
 , /* mNumSlicesForSliceK */ 1
 , /* mNumStages */ 4
-, /* mNumStagesMma */ 2
+, /* mNumStagesMma */ 1
 , /* mNumStagesMmaWithinWorkTile */ 1
-, /* mNumStagesMmaAcrossWorkTile */ 2
+, /* mNumStagesMmaAcrossWorkTile */ 1
 , /* mNumStagesWorkId */ 3
 , /* mOutputDebugTensors */ 0
 , /* mUseShuffledMatrixA */ 1
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
 , /* mTileK */ 512
-, /* mUseUnrollLoop2xForMma */ 1
+, /* mUseUnrollLoop2xForMma */ 0
 , /* mUseCustomMmaSchedule */ 1
 , /* mUseHoistTryWaitForCustomMmaSchedule */ 0
 , /* mUseDeepSeekFp8 */ 0
@@ -768,21 +830,25 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(1)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(1)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
 , /* mBatchedN */ {}
 , /* mBatchMode */ batchedGemm::BatchedGemmOptions::BatchMode(1)
 , /* mNumBatches */ 2
-, /* mIsStaticBatch */ 0
-, /* mNumTokens */ 2
-, /* mRouteAct */ 0
-, /* mGridWaitForPrimaryRouting */ 0
+, /* mIsStaticBatch */ 1
+, /* mNumTokens */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
+, /* mGridWaitForPrimaryRouting */ 1
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin_len, 166912, "BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a", 480, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin_len, 166912, "BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a", 480, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -790,6 +856,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(17826818)
 , /* mDtypeB */ trtllm::gen::Dtype(17826818)
 , /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(17826818)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(17826818)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -806,6 +874,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 64
 , /* mMmaKind */ trtllm::gen::MmaKind(4)
@@ -824,13 +894,11 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseShuffledMatrixA */ 1
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
 , /* mTileK */ 512
-, /* mUseUnrollLoop2xForMma */ 1
+, /* mUseUnrollLoop2xForMma */ 0
 , /* mUseCustomMmaSchedule */ 1
 , /* mUseHoistTryWaitForCustomMmaSchedule */ 0
 , /* mUseDeepSeekFp8 */ 0
@@ -842,6 +910,7 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(1)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(1)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(1)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -850,13 +919,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 0
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_sm100a_cubin, BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_sm100a_cubin_len, 166912, "BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_sm100a", 448, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin_len, 166912, "BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a", 480, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -864,80 +936,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(17826818)
 , /* mDtypeB */ trtllm::gen::Dtype(17826818)
 , /* mDtypeC */ trtllm::gen::Dtype(1052672)
-, /* mEnablesEarlyExit */ 0
-, /* mEnablesDelayedEarlyExit */ 0
-, /* mEnablesGlobalPtxKnobs */ 1
-, /* mEpilogueLdtmDps */ 16
-, /* mEpilogueLdtmBits */ 256
-, /* mEpilogueTileM */ 128
-, /* mEpilogueTileN */ 8
-, /* mGridTriggerSecondaryA */ 0
-, /* mGridTriggerSecondaryB */ 1
-, /* mGridWaitForPrimaryEarlyExit */ 1
-, /* mGridWaitForPrimaryA */ 0
-, /* mGridWaitForPrimaryB */ 1
-, /* mHoistLoadTaskInit */ 1
-, /* mHoistMmaTaskTryWaits */ 0
-, /* mK */ 2048
-, /* mKernelTraits */ {}
-, /* mM */ 256
-, /* mMmaK */ 64
-, /* mMmaKind */ trtllm::gen::MmaKind(4)
-, /* mMmaM */ 128
-, /* mMmaN */ 8
-, /* mMockAllReduce */ 0
-, /* mN */ 256
-, /* mNumSlicesForSplitK */ 1
-, /* mNumSlicesForSliceK */ 1
-, /* mNumStages */ 4
-, /* mNumStagesMma */ 1
-, /* mNumStagesMmaWithinWorkTile */ 1
-, /* mNumStagesMmaAcrossWorkTile */ 1
-, /* mNumStagesWorkId */ 3
-, /* mOutputDebugTensors */ 0
-, /* mUseShuffledMatrixA */ 1
-, /* mSliceK */ 0
-, /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
-, /* mTransposeMmaOutput */ 1
-, /* mTileM */ 128
-, /* mTileN */ 8
-, /* mTileK */ 512
-, /* mUseUnrollLoop2xForMma */ 1
-, /* mUseCustomMmaSchedule */ 1
-, /* mUseHoistTryWaitForCustomMmaSchedule */ 0
-, /* mUseDeepSeekFp8 */ 0
-, /* mUsePerTokenSfA */ 0
-, /* mUsePerTokenSfB */ 0
-, /* mUseTmaStore */ 1
-, /* mUseTwoTmaLoadWarps */ 1
-, /* mUseTwoMmaWarps */ 0
-, /* mSfLayoutA */ trtllm::gen::SfLayout(3)
-, /* mSfLayoutB */ trtllm::gen::SfLayout(1)
-, /* mSfLayoutC */ trtllm::gen::SfLayout(1)
-, /* mTileScheduler */ gemm::TileScheduler(0)
-, /* mActType */ gemmGatedAct::ActType(0)
-, /* mBatchedM */ {}
-, /* mBatchedN */ {}
-, /* mBatchMode */ batchedGemm::BatchedGemmOptions::BatchMode(1)
-, /* mNumBatches */ 2
-, /* mIsStaticBatch */ 1
-, /* mNumTokens */ 0
-, /* mRouteAct */ 0
-, /* mGridWaitForPrimaryRouting */ 1
-, /* mFusedAct */ 0
-, /* mNumRegsPerThreadNonEpilogueWarp */ 0
-, /* mNumRegsPerThreadEpilogueWarp */ 0
- }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E2m1E2m1Fp32_E2m1_Tile128x8x512_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_InplaceRoute_GatedAct_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E2m1E2m1Fp32_E2m1_Tile128x8x512_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_InplaceRoute_GatedAct_dynamic_sm100a_cubin_len, 165888, "BatchedGemmKernel_BatchN_E2m1E2m1Fp32_E2m1_Tile128x8x512_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_InplaceRoute_GatedAct_dynamic_sm100a", 480, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
-, /* mClusterDimX */ 1
-, /* mClusterDimY */ 1
-, /* mClusterDimZ */ 1
-, /* mDtypeAcc */ trtllm::gen::Dtype(1056776)
-, /* mDtypeA */ trtllm::gen::Dtype(17826818)
-, /* mDtypeB */ trtllm::gen::Dtype(17826818)
-, /* mDtypeC */ trtllm::gen::Dtype(17826818)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(17826818)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(17826818)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -954,6 +954,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 64
 , /* mMmaKind */ trtllm::gen::MmaKind(4)
@@ -964,16 +966,14 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumSlicesForSplitK */ 1
 , /* mNumSlicesForSliceK */ 1
 , /* mNumStages */ 4
-, /* mNumStagesMma */ 1
+, /* mNumStagesMma */ 2
 , /* mNumStagesMmaWithinWorkTile */ 1
-, /* mNumStagesMmaAcrossWorkTile */ 1
+, /* mNumStagesMmaAcrossWorkTile */ 2
 , /* mNumStagesWorkId */ 3
 , /* mOutputDebugTensors */ 0
 , /* mUseShuffledMatrixA */ 1
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
@@ -990,29 +990,35 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(1)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(1)
-, /* mTileScheduler */ gemm::TileScheduler(0)
+, /* mSfReshapeFactor */ 1
+, /* mTileScheduler */ gemm::TileScheduler(1)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
 , /* mBatchedN */ {}
 , /* mBatchMode */ batchedGemm::BatchedGemmOptions::BatchMode(1)
-, /* mNumBatches */ 2
+, /* mNumBatches */ 128
 , /* mIsStaticBatch */ 0
-, /* mNumTokens */ 2
-, /* mRouteAct */ 1
-, /* mGridWaitForPrimaryRouting */ 1
-, /* mFusedAct */ 1
+, /* mNumTokens */ 32
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
+, /* mGridWaitForPrimaryRouting */ 0
+, /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E2m1E2m1Fp32_E2m1_Tile128x8x512_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_sm100a_cubin, BatchedGemmKernel_BatchN_E2m1E2m1Fp32_E2m1_Tile128x8x512_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_sm100a_cubin_len, 165888, "BatchedGemmKernel_BatchN_E2m1E2m1Fp32_E2m1_Tile128x8x512_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_sm100a", 448, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin_len, 166912, "BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a", 480, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
 , /* mDtypeAcc */ trtllm::gen::Dtype(1056776)
 , /* mDtypeA */ trtllm::gen::Dtype(17826818)
 , /* mDtypeB */ trtllm::gen::Dtype(17826818)
-, /* mDtypeC */ trtllm::gen::Dtype(17826818)
-, /* mEnablesEarlyExit */ 0
+, /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(17826818)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(17826818)
+, /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
 , /* mEpilogueLdtmDps */ 16
@@ -1028,6 +1034,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 64
 , /* mMmaKind */ trtllm::gen::MmaKind(4)
@@ -1038,16 +1046,14 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumSlicesForSplitK */ 1
 , /* mNumSlicesForSliceK */ 1
 , /* mNumStages */ 4
-, /* mNumStagesMma */ 1
+, /* mNumStagesMma */ 2
 , /* mNumStagesMmaWithinWorkTile */ 1
-, /* mNumStagesMmaAcrossWorkTile */ 1
+, /* mNumStagesMmaAcrossWorkTile */ 2
 , /* mNumStagesWorkId */ 3
 , /* mOutputDebugTensors */ 0
 , /* mUseShuffledMatrixA */ 1
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
@@ -1064,28 +1070,34 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(1)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(1)
-, /* mTileScheduler */ gemm::TileScheduler(0)
+, /* mSfReshapeFactor */ 1
+, /* mTileScheduler */ gemm::TileScheduler(1)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
 , /* mBatchedN */ {}
 , /* mBatchMode */ batchedGemm::BatchedGemmOptions::BatchMode(1)
-, /* mNumBatches */ 2
-, /* mIsStaticBatch */ 1
-, /* mNumTokens */ 0
-, /* mRouteAct */ 0
-, /* mGridWaitForPrimaryRouting */ 1
+, /* mNumBatches */ 128
+, /* mIsStaticBatch */ 0
+, /* mNumTokens */ 32
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
+, /* mGridWaitForPrimaryRouting */ 0
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E2m1E2m1Fp32_E2m1_Tile128x8x512u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_InplaceRoute_GatedAct_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E2m1E2m1Fp32_E2m1_Tile128x8x512u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_InplaceRoute_GatedAct_dynamic_sm100a_cubin_len, 165888, "BatchedGemmKernel_BatchN_E2m1E2m1Fp32_E2m1_Tile128x8x512u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_InplaceRoute_GatedAct_dynamic_sm100a", 480, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin_len, 166912, "BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a", 448, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
 , /* mDtypeAcc */ trtllm::gen::Dtype(1056776)
 , /* mDtypeA */ trtllm::gen::Dtype(17826818)
 , /* mDtypeB */ trtllm::gen::Dtype(17826818)
-, /* mDtypeC */ trtllm::gen::Dtype(17826818)
+, /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(17826818)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(17826818)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -1102,6 +1114,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 64
 , /* mMmaKind */ trtllm::gen::MmaKind(4)
@@ -1120,8 +1134,6 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseShuffledMatrixA */ 1
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
@@ -1138,6 +1150,7 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(1)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(1)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -1146,94 +1159,25 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 1
-, /* mGridWaitForPrimaryRouting */ 1
-, /* mFusedAct */ 1
-, /* mNumRegsPerThreadNonEpilogueWarp */ 0
-, /* mNumRegsPerThreadEpilogueWarp */ 0
- }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E2m1E2m1Fp32_E2m1_Tile128x8x512u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_sm100a_cubin, BatchedGemmKernel_BatchN_E2m1E2m1Fp32_E2m1_Tile128x8x512u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_sm100a_cubin_len, 165888, "BatchedGemmKernel_BatchN_E2m1E2m1Fp32_E2m1_Tile128x8x512u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_sm100a", 448, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
-, /* mClusterDimX */ 1
-, /* mClusterDimY */ 1
-, /* mClusterDimZ */ 1
-, /* mDtypeAcc */ trtllm::gen::Dtype(1056776)
-, /* mDtypeA */ trtllm::gen::Dtype(17826818)
-, /* mDtypeB */ trtllm::gen::Dtype(17826818)
-, /* mDtypeC */ trtllm::gen::Dtype(17826818)
-, /* mEnablesEarlyExit */ 0
-, /* mEnablesDelayedEarlyExit */ 0
-, /* mEnablesGlobalPtxKnobs */ 1
-, /* mEpilogueLdtmDps */ 16
-, /* mEpilogueLdtmBits */ 256
-, /* mEpilogueTileM */ 128
-, /* mEpilogueTileN */ 8
-, /* mGridTriggerSecondaryA */ 0
-, /* mGridTriggerSecondaryB */ 1
-, /* mGridWaitForPrimaryEarlyExit */ 1
-, /* mGridWaitForPrimaryA */ 0
-, /* mGridWaitForPrimaryB */ 1
-, /* mHoistLoadTaskInit */ 1
-, /* mHoistMmaTaskTryWaits */ 0
-, /* mK */ 2048
-, /* mKernelTraits */ {}
-, /* mM */ 256
-, /* mMmaK */ 64
-, /* mMmaKind */ trtllm::gen::MmaKind(4)
-, /* mMmaM */ 128
-, /* mMmaN */ 8
-, /* mMockAllReduce */ 0
-, /* mN */ 256
-, /* mNumSlicesForSplitK */ 1
-, /* mNumSlicesForSliceK */ 1
-, /* mNumStages */ 4
-, /* mNumStagesMma */ 1
-, /* mNumStagesMmaWithinWorkTile */ 1
-, /* mNumStagesMmaAcrossWorkTile */ 1
-, /* mNumStagesWorkId */ 3
-, /* mOutputDebugTensors */ 0
-, /* mUseShuffledMatrixA */ 1
-, /* mSliceK */ 0
-, /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
-, /* mTransposeMmaOutput */ 1
-, /* mTileM */ 128
-, /* mTileN */ 8
-, /* mTileK */ 512
-, /* mUseUnrollLoop2xForMma */ 1
-, /* mUseCustomMmaSchedule */ 1
-, /* mUseHoistTryWaitForCustomMmaSchedule */ 0
-, /* mUseDeepSeekFp8 */ 0
-, /* mUsePerTokenSfA */ 0
-, /* mUsePerTokenSfB */ 0
-, /* mUseTmaStore */ 1
-, /* mUseTwoTmaLoadWarps */ 1
-, /* mUseTwoMmaWarps */ 0
-, /* mSfLayoutA */ trtllm::gen::SfLayout(3)
-, /* mSfLayoutB */ trtllm::gen::SfLayout(1)
-, /* mSfLayoutC */ trtllm::gen::SfLayout(1)
-, /* mTileScheduler */ gemm::TileScheduler(0)
-, /* mActType */ gemmGatedAct::ActType(0)
-, /* mBatchedM */ {}
-, /* mBatchedN */ {}
-, /* mBatchMode */ batchedGemm::BatchedGemmOptions::BatchMode(1)
-, /* mNumBatches */ 2
-, /* mIsStaticBatch */ 1
-, /* mNumTokens */ 0
-, /* mRouteAct */ 0
-, /* mGridWaitForPrimaryRouting */ 1
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
+, /* mGridWaitForPrimaryRouting */ 0
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Fp16_Tile128x8x512_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_sm100a_cubin, BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Fp16_Tile128x8x512_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_sm100a_cubin_len, 166912, "BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Fp16_Tile128x8x512_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_sm100a", 448, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin, BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin_len, 166912, "BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a", 448, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
 , /* mDtypeAcc */ trtllm::gen::Dtype(1056776)
 , /* mDtypeA */ trtllm::gen::Dtype(17826818)
 , /* mDtypeB */ trtllm::gen::Dtype(17826818)
-, /* mDtypeC */ trtllm::gen::Dtype(1052679)
+, /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(17826818)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(17826818)
 , /* mEnablesEarlyExit */ 0
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -1250,6 +1194,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 64
 , /* mMmaKind */ trtllm::gen::MmaKind(4)
@@ -1268,13 +1214,11 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseShuffledMatrixA */ 1
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
 , /* mTileK */ 512
-, /* mUseUnrollLoop2xForMma */ 0
+, /* mUseUnrollLoop2xForMma */ 1
 , /* mUseCustomMmaSchedule */ 1
 , /* mUseHoistTryWaitForCustomMmaSchedule */ 0
 , /* mUseDeepSeekFp8 */ 0
@@ -1286,6 +1230,7 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(1)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(1)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -1294,21 +1239,26 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 1
 , /* mNumTokens */ 0
-, /* mRouteAct */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 1
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Fp16_Tile128x8x512u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_sm100a_cubin, BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Fp16_Tile128x8x512u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_sm100a_cubin_len, 166912, "BatchedGemmKernel_BatchN_E2m1E2m1Fp32_Fp16_Tile128x8x512u2_EpiTile128x8_Mma128x8x64_Cluster1x1x1_transposeMmaOutput_sm100a", 448, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin_len, 166912, "BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a", 480, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
 , /* mDtypeAcc */ trtllm::gen::Dtype(1056776)
 , /* mDtypeA */ trtllm::gen::Dtype(17826818)
 , /* mDtypeB */ trtllm::gen::Dtype(17826818)
-, /* mDtypeC */ trtllm::gen::Dtype(1052679)
-, /* mEnablesEarlyExit */ 0
+, /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(17826818)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(17826818)
+, /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
 , /* mEpilogueLdtmDps */ 16
@@ -1324,6 +1274,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 64
 , /* mMmaKind */ trtllm::gen::MmaKind(4)
@@ -1334,16 +1286,14 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumSlicesForSplitK */ 1
 , /* mNumSlicesForSliceK */ 1
 , /* mNumStages */ 4
-, /* mNumStagesMma */ 1
+, /* mNumStagesMma */ 2
 , /* mNumStagesMmaWithinWorkTile */ 1
-, /* mNumStagesMmaAcrossWorkTile */ 1
+, /* mNumStagesMmaAcrossWorkTile */ 2
 , /* mNumStagesWorkId */ 3
 , /* mOutputDebugTensors */ 0
 , /* mUseShuffledMatrixA */ 1
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
@@ -1360,21 +1310,25 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(1)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(1)
-, /* mTileScheduler */ gemm::TileScheduler(0)
+, /* mSfReshapeFactor */ 1
+, /* mTileScheduler */ gemm::TileScheduler(1)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
 , /* mBatchedN */ {}
 , /* mBatchMode */ batchedGemm::BatchedGemmOptions::BatchMode(1)
 , /* mNumBatches */ 2
-, /* mIsStaticBatch */ 1
-, /* mNumTokens */ 0
-, /* mRouteAct */ 0
-, /* mGridWaitForPrimaryRouting */ 1
+, /* mIsStaticBatch */ 0
+, /* mNumTokens */ 2
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
+, /* mGridWaitForPrimaryRouting */ 0
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x128x128_EpiTile64x128_Mma64x128x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x128x128_EpiTile64x128_Mma64x128x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin_len, 175104, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x128x128_EpiTile64x128_Mma64x128x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a", 384, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x128x128_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x128x128_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin_len, 175104, "BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x128x128_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a", 384, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -1382,6 +1336,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
 , /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -1398,6 +1354,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 1
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
@@ -1416,8 +1374,6 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseShuffledMatrixA */ 0
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 128
@@ -1434,6 +1390,7 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -1442,13 +1399,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 0
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x128x128_EpiTile64x128_Mma64x128x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x128x128_EpiTile64x128_Mma64x128x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin_len, 175104, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x128x128_EpiTile64x128_Mma64x128x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x128x128_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x128x128_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin_len, 175104, "BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x128x128_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -1456,6 +1416,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
 , /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -1472,6 +1434,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 1
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
@@ -1490,8 +1454,6 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseShuffledMatrixA */ 0
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 128
@@ -1508,6 +1470,7 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(1)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -1516,13 +1479,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 0
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x128x128u2_EpiTile64x128_Mma64x128x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x128x128u2_EpiTile64x128_Mma64x128x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin_len, 175104, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x128x128u2_EpiTile64x128_Mma64x128x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a", 384, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x128x128u2_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x128x128u2_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin_len, 175104, "BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x128x128u2_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a", 384, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -1530,6 +1496,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
 , /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -1546,6 +1514,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 1
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
@@ -1564,8 +1534,6 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseShuffledMatrixA */ 0
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 128
@@ -1582,6 +1550,7 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -1590,13 +1559,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 0
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x128x128u2_EpiTile64x128_Mma64x128x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x128x128u2_EpiTile64x128_Mma64x128x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin_len, 175104, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x128x128u2_EpiTile64x128_Mma64x128x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x128x128u2_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x128x128u2_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin_len, 175104, "BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x128x128u2_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -1604,6 +1576,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
 , /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -1620,6 +1594,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 1
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
@@ -1638,8 +1614,6 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseShuffledMatrixA */ 0
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 128
@@ -1656,6 +1630,7 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(1)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -1664,13 +1639,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 0
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x16x128_EpiTile64x16_Mma64x16x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x16x128_EpiTile64x16_Mma64x16x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin_len, 84992, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x16x128_EpiTile64x16_Mma64x16x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a", 384, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x16x128_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x16x128_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin_len, 84992, "BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x16x128_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a", 384, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -1678,6 +1656,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
 , /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -1694,6 +1674,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 1
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
@@ -1712,8 +1694,6 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseShuffledMatrixA */ 0
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 16
@@ -1730,6 +1710,7 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -1738,13 +1719,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 0
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x16x128_EpiTile64x16_Mma64x16x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x16x128_EpiTile64x16_Mma64x16x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin_len, 84992, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x16x128_EpiTile64x16_Mma64x16x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x16x128_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x16x128_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin_len, 84992, "BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x16x128_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -1752,6 +1736,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
 , /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -1768,6 +1754,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 1
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
@@ -1786,8 +1774,6 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseShuffledMatrixA */ 0
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 16
@@ -1804,6 +1790,7 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(1)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -1812,13 +1799,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 0
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x16x128u2_EpiTile64x16_Mma64x16x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x16x128u2_EpiTile64x16_Mma64x16x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin_len, 84992, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x16x128u2_EpiTile64x16_Mma64x16x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a", 384, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x16x128u2_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x16x128u2_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin_len, 84992, "BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x16x128u2_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a", 384, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -1826,6 +1816,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
 , /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -1842,6 +1834,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 1
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
@@ -1860,8 +1854,6 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseShuffledMatrixA */ 0
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 16
@@ -1878,6 +1870,7 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -1886,13 +1879,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 0
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x16x128u2_EpiTile64x16_Mma64x16x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x16x128u2_EpiTile64x16_Mma64x16x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin_len, 84992, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x16x128u2_EpiTile64x16_Mma64x16x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x16x128u2_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x16x128u2_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin_len, 84992, "BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x16x128u2_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -1900,6 +1896,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
 , /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -1916,6 +1914,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 1
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
@@ -1934,8 +1934,6 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseShuffledMatrixA */ 0
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 16
@@ -1952,6 +1950,7 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(1)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -1960,13 +1959,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 0
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x32x128_EpiTile64x32_Mma64x32x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x32x128_EpiTile64x32_Mma64x32x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin_len, 97280, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x32x128_EpiTile64x32_Mma64x32x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a", 384, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x32x128_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x32x128_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin_len, 97280, "BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x32x128_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a", 384, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -1974,6 +1976,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
 , /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -1990,6 +1994,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 1
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
@@ -2008,8 +2014,6 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseShuffledMatrixA */ 0
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 32
@@ -2026,6 +2030,7 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -2034,13 +2039,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 0
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x32x128_EpiTile64x32_Mma64x32x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x32x128_EpiTile64x32_Mma64x32x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin_len, 97280, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x32x128_EpiTile64x32_Mma64x32x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x32x128_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x32x128_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin_len, 97280, "BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x32x128_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -2048,6 +2056,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
 , /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -2064,6 +2074,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 1
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
@@ -2082,8 +2094,6 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseShuffledMatrixA */ 0
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 32
@@ -2100,6 +2110,7 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(1)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -2108,13 +2119,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 0
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x32x128u2_EpiTile64x32_Mma64x32x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x32x128u2_EpiTile64x32_Mma64x32x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin_len, 97280, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x32x128u2_EpiTile64x32_Mma64x32x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a", 384, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x32x128u2_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x32x128u2_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin_len, 97280, "BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x32x128u2_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a", 384, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -2122,6 +2136,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
 , /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -2138,6 +2154,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 1
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
@@ -2156,8 +2174,6 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseShuffledMatrixA */ 0
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 32
@@ -2174,6 +2190,7 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -2182,13 +2199,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 0
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x32x128u2_EpiTile64x32_Mma64x32x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x32x128u2_EpiTile64x32_Mma64x32x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin_len, 97280, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x32x128u2_EpiTile64x32_Mma64x32x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x32x128u2_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x32x128u2_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin_len, 97280, "BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x32x128u2_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -2196,6 +2216,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
 , /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -2212,6 +2234,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 1
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
@@ -2230,8 +2254,6 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseShuffledMatrixA */ 0
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 32
@@ -2248,6 +2270,7 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(1)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -2256,13 +2279,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 0
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x64x128_EpiTile64x64_Mma64x64x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x64x128_EpiTile64x64_Mma64x64x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin_len, 123904, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x64x128_EpiTile64x64_Mma64x64x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a", 384, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x64x128_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x64x128_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin_len, 123904, "BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x64x128_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a", 384, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -2270,6 +2296,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
 , /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -2286,6 +2314,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 1
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
@@ -2304,8 +2334,6 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseShuffledMatrixA */ 0
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 64
@@ -2322,6 +2350,7 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -2330,13 +2359,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 0
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x64x128_EpiTile64x64_Mma64x64x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x64x128_EpiTile64x64_Mma64x64x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin_len, 123904, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x64x128_EpiTile64x64_Mma64x64x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x64x128_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x64x128_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin_len, 123904, "BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x64x128_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -2344,6 +2376,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
 , /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -2360,6 +2394,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 1
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
@@ -2378,8 +2414,6 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseShuffledMatrixA */ 0
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 64
@@ -2396,6 +2430,7 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(1)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -2404,13 +2439,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 0
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x64x128u2_EpiTile64x64_Mma64x64x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x64x128u2_EpiTile64x64_Mma64x64x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin_len, 123904, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x64x128u2_EpiTile64x64_Mma64x64x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a", 384, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x64x128u2_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x64x128u2_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin_len, 123904, "BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x64x128u2_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a", 384, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -2418,6 +2456,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
 , /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -2434,6 +2474,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 1
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
@@ -2452,8 +2494,6 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseShuffledMatrixA */ 0
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 64
@@ -2470,6 +2510,7 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -2478,13 +2519,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 0
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x64x128u2_EpiTile64x64_Mma64x64x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x64x128u2_EpiTile64x64_Mma64x64x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin_len, 123904, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x64x128u2_EpiTile64x64_Mma64x64x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x64x128u2_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x64x128u2_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin_len, 123904, "BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x64x128u2_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -2492,6 +2536,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
 , /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -2508,6 +2554,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 1
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
@@ -2526,8 +2574,6 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseShuffledMatrixA */ 0
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 64
@@ -2544,6 +2590,7 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(1)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -2552,13 +2599,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 0
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x128_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x128_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin_len, 149504, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x128_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a", 384, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin_len, 149504, "BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a", 384, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -2566,6 +2616,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
 , /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -2582,6 +2634,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
@@ -2592,16 +2646,14 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumSlicesForSplitK */ 1
 , /* mNumSlicesForSliceK */ 1
 , /* mNumStages */ 8
-, /* mNumStagesMma */ 4
+, /* mNumStagesMma */ 2
 , /* mNumStagesMmaWithinWorkTile */ 2
-, /* mNumStagesMmaAcrossWorkTile */ 2
+, /* mNumStagesMmaAcrossWorkTile */ 1
 , /* mNumStagesWorkId */ 3
 , /* mOutputDebugTensors */ 0
 , /* mUseShuffledMatrixA */ 0
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
@@ -2618,6 +2670,7 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -2626,13 +2679,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 0
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x128_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x128_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin_len, 149504, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x128_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin, BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin_len, 61440, "BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a", 384, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -2640,81 +2696,9 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
 , /* mDtypeC */ trtllm::gen::Dtype(1052672)
-, /* mEnablesEarlyExit */ 1
-, /* mEnablesDelayedEarlyExit */ 0
-, /* mEnablesGlobalPtxKnobs */ 1
-, /* mEpilogueLdtmDps */ 16
-, /* mEpilogueLdtmBits */ 256
-, /* mEpilogueTileM */ 64
-, /* mEpilogueTileN */ 8
-, /* mGridTriggerSecondaryA */ 0
-, /* mGridTriggerSecondaryB */ 1
-, /* mGridWaitForPrimaryEarlyExit */ 1
-, /* mGridWaitForPrimaryA */ 0
-, /* mGridWaitForPrimaryB */ 1
-, /* mHoistLoadTaskInit */ 1
-, /* mHoistMmaTaskTryWaits */ 0
-, /* mK */ 2048
-, /* mKernelTraits */ {}
-, /* mM */ 256
-, /* mMmaK */ 32
-, /* mMmaKind */ trtllm::gen::MmaKind(2)
-, /* mMmaM */ 64
-, /* mMmaN */ 8
-, /* mMockAllReduce */ 0
-, /* mN */ 256
-, /* mNumSlicesForSplitK */ 1
-, /* mNumSlicesForSliceK */ 1
-, /* mNumStages */ 8
-, /* mNumStagesMma */ 4
-, /* mNumStagesMmaWithinWorkTile */ 2
-, /* mNumStagesMmaAcrossWorkTile */ 2
-, /* mNumStagesWorkId */ 3
-, /* mOutputDebugTensors */ 0
-, /* mUseShuffledMatrixA */ 0
-, /* mSliceK */ 0
-, /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
-, /* mTransposeMmaOutput */ 1
-, /* mTileM */ 128
-, /* mTileN */ 8
-, /* mTileK */ 128
-, /* mUseUnrollLoop2xForMma */ 0
-, /* mUseCustomMmaSchedule */ 1
-, /* mUseHoistTryWaitForCustomMmaSchedule */ 0
-, /* mUseDeepSeekFp8 */ 1
-, /* mUsePerTokenSfA */ 0
-, /* mUsePerTokenSfB */ 0
-, /* mUseTmaStore */ 1
-, /* mUseTwoTmaLoadWarps */ 1
-, /* mUseTwoMmaWarps */ 0
-, /* mSfLayoutA */ trtllm::gen::SfLayout(3)
-, /* mSfLayoutB */ trtllm::gen::SfLayout(3)
-, /* mSfLayoutC */ trtllm::gen::SfLayout(3)
-, /* mTileScheduler */ gemm::TileScheduler(1)
-, /* mActType */ gemmGatedAct::ActType(0)
-, /* mBatchedM */ {}
-, /* mBatchedN */ {}
-, /* mBatchMode */ batchedGemm::BatchedGemmOptions::BatchMode(1)
-, /* mNumBatches */ 2
-, /* mIsStaticBatch */ 0
-, /* mNumTokens */ 2
-, /* mRouteAct */ 0
-, /* mGridWaitForPrimaryRouting */ 0
-, /* mFusedAct */ 0
-, /* mNumRegsPerThreadNonEpilogueWarp */ 0
-, /* mNumRegsPerThreadEpilogueWarp */ 0
- }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x128_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x128_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_sm100a_cubin_len, 61440, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x128_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_sm100a", 384, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
-, /* mClusterDimX */ 1
-, /* mClusterDimY */ 1
-, /* mClusterDimZ */ 1
-, /* mDtypeAcc */ trtllm::gen::Dtype(1056776)
-, /* mDtypeA */ trtllm::gen::Dtype(1050629)
-, /* mDtypeB */ trtllm::gen::Dtype(1050629)
-, /* mDtypeC */ trtllm::gen::Dtype(1052672)
-, /* mEnablesEarlyExit */ 0
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
+, /* mEnablesEarlyExit */ 0
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
 , /* mEpilogueLdtmDps */ 16
@@ -2730,6 +2714,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
@@ -2748,8 +2734,6 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseShuffledMatrixA */ 0
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
@@ -2766,6 +2750,7 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -2774,13 +2759,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 1
 , /* mNumTokens */ 0
-, /* mRouteAct */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 1
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x128u2_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x128u2_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a_cubin_len, 149504, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x128u2_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_dynamic_sm100a", 384, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin_len, 149504, "BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -2788,6 +2776,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
 , /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -2804,6 +2794,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
@@ -2822,13 +2814,11 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseShuffledMatrixA */ 0
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
 , /* mTileK */ 128
-, /* mUseUnrollLoop2xForMma */ 1
+, /* mUseUnrollLoop2xForMma */ 0
 , /* mUseCustomMmaSchedule */ 1
 , /* mUseHoistTryWaitForCustomMmaSchedule */ 0
 , /* mUseDeepSeekFp8 */ 1
@@ -2840,7 +2830,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
-, /* mTileScheduler */ gemm::TileScheduler(0)
+, /* mSfReshapeFactor */ 1
+, /* mTileScheduler */ gemm::TileScheduler(1)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
 , /* mBatchedN */ {}
@@ -2848,13 +2839,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 0
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x128u2_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x128u2_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a_cubin_len, 149504, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x128u2_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_persistent_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin_len, 149504, "BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a", 384, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -2862,6 +2856,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
 , /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -2878,6 +2874,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
@@ -2888,16 +2886,14 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumSlicesForSplitK */ 1
 , /* mNumSlicesForSliceK */ 1
 , /* mNumStages */ 8
-, /* mNumStagesMma */ 4
+, /* mNumStagesMma */ 2
 , /* mNumStagesMmaWithinWorkTile */ 2
-, /* mNumStagesMmaAcrossWorkTile */ 2
+, /* mNumStagesMmaAcrossWorkTile */ 1
 , /* mNumStagesWorkId */ 3
 , /* mOutputDebugTensors */ 0
 , /* mUseShuffledMatrixA */ 0
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
@@ -2914,7 +2910,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
-, /* mTileScheduler */ gemm::TileScheduler(1)
+, /* mSfReshapeFactor */ 1
+, /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
 , /* mBatchedN */ {}
@@ -2922,13 +2919,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 0
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x128u2_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x128u2_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_sm100a_cubin_len, 61440, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x128u2_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_sm100a", 384, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin, BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin_len, 61440, "BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a", 384, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -2936,6 +2936,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
 , /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
 , /* mEnablesEarlyExit */ 0
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -2952,6 +2954,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
@@ -2970,8 +2974,6 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseShuffledMatrixA */ 0
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
@@ -2988,6 +2990,7 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -2996,13 +2999,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 1
 , /* mNumTokens */ 0
-, /* mRouteAct */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 1
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin_len, 217088, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a", 224, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin_len, 149504, "BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -3010,12 +3016,14 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
 , /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
 , /* mEpilogueLdtmDps */ 16
 , /* mEpilogueLdtmBits */ 256
-, /* mEpilogueTileM */ 128
+, /* mEpilogueTileM */ 64
 , /* mEpilogueTileN */ 8
 , /* mGridTriggerSecondaryA */ 0
 , /* mGridTriggerSecondaryB */ 1
@@ -3026,34 +3034,34 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
-, /* mMmaM */ 128
+, /* mMmaM */ 64
 , /* mMmaN */ 8
 , /* mMockAllReduce */ 0
 , /* mN */ 256
 , /* mNumSlicesForSplitK */ 1
 , /* mNumSlicesForSliceK */ 1
-, /* mNumStages */ 3
-, /* mNumStagesMma */ 2
-, /* mNumStagesMmaWithinWorkTile */ 1
+, /* mNumStages */ 8
+, /* mNumStagesMma */ 4
+, /* mNumStagesMmaWithinWorkTile */ 2
 , /* mNumStagesMmaAcrossWorkTile */ 2
 , /* mNumStagesWorkId */ 3
 , /* mOutputDebugTensors */ 0
-, /* mUseShuffledMatrixA */ 1
+, /* mUseShuffledMatrixA */ 0
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
-, /* mTileK */ 512
-, /* mUseUnrollLoop2xForMma */ 0
+, /* mTileK */ 128
+, /* mUseUnrollLoop2xForMma */ 1
 , /* mUseCustomMmaSchedule */ 1
 , /* mUseHoistTryWaitForCustomMmaSchedule */ 0
-, /* mUseDeepSeekFp8 */ 0
+, /* mUseDeepSeekFp8 */ 1
 , /* mUsePerTokenSfA */ 0
 , /* mUsePerTokenSfB */ 0
 , /* mUseTmaStore */ 1
@@ -3062,7 +3070,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
-, /* mTileScheduler */ gemm::TileScheduler(0)
+, /* mSfReshapeFactor */ 1
+, /* mTileScheduler */ gemm::TileScheduler(1)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
 , /* mBatchedN */ {}
@@ -3070,13 +3079,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 0
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin_len, 217088, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a", 256, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin_len, 217088, "BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a", 224, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -3084,6 +3096,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
 , /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -3100,6 +3114,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
@@ -3110,16 +3126,14 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumSlicesForSplitK */ 1
 , /* mNumSlicesForSliceK */ 1
 , /* mNumStages */ 3
-, /* mNumStagesMma */ 2
+, /* mNumStagesMma */ 1
 , /* mNumStagesMmaWithinWorkTile */ 1
-, /* mNumStagesMmaAcrossWorkTile */ 2
+, /* mNumStagesMmaAcrossWorkTile */ 1
 , /* mNumStagesWorkId */ 3
 , /* mOutputDebugTensors */ 0
 , /* mUseShuffledMatrixA */ 1
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
@@ -3136,7 +3150,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
-, /* mTileScheduler */ gemm::TileScheduler(1)
+, /* mSfReshapeFactor */ 1
+, /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
 , /* mBatchedN */ {}
@@ -3144,13 +3159,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 0
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_sm100a_cubin_len, 217088, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_sm100a", 224, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin, BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin_len, 217088, "BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a", 224, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -3158,6 +3176,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
 , /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
 , /* mEnablesEarlyExit */ 0
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -3174,6 +3194,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
@@ -3192,8 +3214,6 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseShuffledMatrixA */ 1
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
@@ -3210,6 +3230,7 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -3218,13 +3239,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 1
 , /* mNumTokens */ 0
-, /* mRouteAct */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 1
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin_len, 217088, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a", 224, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin_len, 217088, "BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a", 256, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -3232,6 +3256,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
 , /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -3248,6 +3274,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
@@ -3266,13 +3294,11 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseShuffledMatrixA */ 1
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
 , /* mTileK */ 512
-, /* mUseUnrollLoop2xForMma */ 1
+, /* mUseUnrollLoop2xForMma */ 0
 , /* mUseCustomMmaSchedule */ 1
 , /* mUseHoistTryWaitForCustomMmaSchedule */ 0
 , /* mUseDeepSeekFp8 */ 0
@@ -3284,7 +3310,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
-, /* mTileScheduler */ gemm::TileScheduler(0)
+, /* mSfReshapeFactor */ 1
+, /* mTileScheduler */ gemm::TileScheduler(1)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
 , /* mBatchedN */ {}
@@ -3292,13 +3319,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 0
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin_len, 217088, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a", 256, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin_len, 217088, "BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a", 224, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -3306,6 +3336,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
 , /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -3322,6 +3354,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
@@ -3332,16 +3366,14 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumSlicesForSplitK */ 1
 , /* mNumSlicesForSliceK */ 1
 , /* mNumStages */ 3
-, /* mNumStagesMma */ 2
+, /* mNumStagesMma */ 1
 , /* mNumStagesMmaWithinWorkTile */ 1
-, /* mNumStagesMmaAcrossWorkTile */ 2
+, /* mNumStagesMmaAcrossWorkTile */ 1
 , /* mNumStagesWorkId */ 3
 , /* mOutputDebugTensors */ 0
 , /* mUseShuffledMatrixA */ 1
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
@@ -3358,7 +3390,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
-, /* mTileScheduler */ gemm::TileScheduler(1)
+, /* mSfReshapeFactor */ 1
+, /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
 , /* mBatchedN */ {}
@@ -3366,13 +3399,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 0
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_sm100a_cubin_len, 217088, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_sm100a", 224, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin, BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin_len, 217088, "BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a", 224, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -3380,6 +3416,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
 , /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
 , /* mEnablesEarlyExit */ 0
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -3396,6 +3434,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
@@ -3414,8 +3454,6 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseShuffledMatrixA */ 1
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
@@ -3432,6 +3470,7 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -3440,64 +3479,69 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 1
 , /* mNumTokens */ 0
-, /* mRouteAct */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 1
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x128x128_EpiTile64x128_Mma64x128x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x128x128_EpiTile64x128_Mma64x128x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin_len, 158720, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x128x128_EpiTile64x128_Mma64x128x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin_len, 217088, "BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a", 256, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
 , /* mDtypeAcc */ trtllm::gen::Dtype(1056776)
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
-, /* mDtypeC */ trtllm::gen::Dtype(1050629)
+, /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
 , /* mEpilogueLdtmDps */ 16
 , /* mEpilogueLdtmBits */ 256
-, /* mEpilogueTileM */ 64
-, /* mEpilogueTileN */ 128
+, /* mEpilogueTileM */ 128
+, /* mEpilogueTileN */ 8
 , /* mGridTriggerSecondaryA */ 0
 , /* mGridTriggerSecondaryB */ 1
 , /* mGridWaitForPrimaryEarlyExit */ 1
 , /* mGridWaitForPrimaryA */ 0
 , /* mGridWaitForPrimaryB */ 1
 , /* mHoistLoadTaskInit */ 1
-, /* mHoistMmaTaskTryWaits */ 1
+, /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
-, /* mMmaM */ 64
-, /* mMmaN */ 128
+, /* mMmaM */ 128
+, /* mMmaN */ 8
 , /* mMockAllReduce */ 0
 , /* mN */ 256
 , /* mNumSlicesForSplitK */ 1
 , /* mNumSlicesForSliceK */ 1
-, /* mNumStages */ 4
-, /* mNumStagesMma */ 4
-, /* mNumStagesMmaWithinWorkTile */ 2
+, /* mNumStages */ 3
+, /* mNumStagesMma */ 2
+, /* mNumStagesMmaWithinWorkTile */ 1
 , /* mNumStagesMmaAcrossWorkTile */ 2
 , /* mNumStagesWorkId */ 3
 , /* mOutputDebugTensors */ 0
-, /* mUseShuffledMatrixA */ 0
+, /* mUseShuffledMatrixA */ 1
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
-, /* mTileN */ 128
-, /* mTileK */ 128
-, /* mUseUnrollLoop2xForMma */ 0
+, /* mTileN */ 8
+, /* mTileK */ 512
+, /* mUseUnrollLoop2xForMma */ 1
 , /* mUseCustomMmaSchedule */ 1
 , /* mUseHoistTryWaitForCustomMmaSchedule */ 0
-, /* mUseDeepSeekFp8 */ 1
+, /* mUseDeepSeekFp8 */ 0
 , /* mUsePerTokenSfA */ 0
 , /* mUsePerTokenSfB */ 0
 , /* mUseTmaStore */ 1
@@ -3506,7 +3550,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
-, /* mTileScheduler */ gemm::TileScheduler(0)
+, /* mSfReshapeFactor */ 1
+, /* mTileScheduler */ gemm::TileScheduler(1)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
 , /* mBatchedN */ {}
@@ -3514,72 +3559,78 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 1
-, /* mGridWaitForPrimaryRouting */ 1
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
+, /* mGridWaitForPrimaryRouting */ 0
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x128x128u2_EpiTile64x128_Mma64x128x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x128x128u2_EpiTile64x128_Mma64x128x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin_len, 158720, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x128x128u2_EpiTile64x128_Mma64x128x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin_len, 153600, "BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a", 448, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
 , /* mDtypeAcc */ trtllm::gen::Dtype(1056776)
-, /* mDtypeA */ trtllm::gen::Dtype(1050629)
-, /* mDtypeB */ trtllm::gen::Dtype(1050629)
-, /* mDtypeC */ trtllm::gen::Dtype(1050629)
+, /* mDtypeA */ trtllm::gen::Dtype(17826828)
+, /* mDtypeB */ trtllm::gen::Dtype(17827853)
+, /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(17826828)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(17827853)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
 , /* mEpilogueLdtmDps */ 16
 , /* mEpilogueLdtmBits */ 256
-, /* mEpilogueTileM */ 64
-, /* mEpilogueTileN */ 128
+, /* mEpilogueTileM */ 128
+, /* mEpilogueTileN */ 8
 , /* mGridTriggerSecondaryA */ 0
 , /* mGridTriggerSecondaryB */ 1
 , /* mGridWaitForPrimaryEarlyExit */ 1
 , /* mGridWaitForPrimaryA */ 0
 , /* mGridWaitForPrimaryB */ 1
 , /* mHoistLoadTaskInit */ 1
-, /* mHoistMmaTaskTryWaits */ 1
+, /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
-, /* mMmaKind */ trtllm::gen::MmaKind(2)
-, /* mMmaM */ 64
-, /* mMmaN */ 128
+, /* mMmaKind */ trtllm::gen::MmaKind(5)
+, /* mMmaM */ 128
+, /* mMmaN */ 8
 , /* mMockAllReduce */ 0
 , /* mN */ 256
 , /* mNumSlicesForSplitK */ 1
 , /* mNumSlicesForSliceK */ 1
 , /* mNumStages */ 4
-, /* mNumStagesMma */ 4
-, /* mNumStagesMmaWithinWorkTile */ 2
-, /* mNumStagesMmaAcrossWorkTile */ 2
+, /* mNumStagesMma */ 1
+, /* mNumStagesMmaWithinWorkTile */ 1
+, /* mNumStagesMmaAcrossWorkTile */ 1
 , /* mNumStagesWorkId */ 3
 , /* mOutputDebugTensors */ 0
-, /* mUseShuffledMatrixA */ 0
+, /* mUseShuffledMatrixA */ 1
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
-, /* mTileN */ 128
-, /* mTileK */ 128
-, /* mUseUnrollLoop2xForMma */ 1
+, /* mTileN */ 8
+, /* mTileK */ 256
+, /* mUseUnrollLoop2xForMma */ 0
 , /* mUseCustomMmaSchedule */ 1
 , /* mUseHoistTryWaitForCustomMmaSchedule */ 0
-, /* mUseDeepSeekFp8 */ 1
+, /* mUseDeepSeekFp8 */ 0
 , /* mUsePerTokenSfA */ 0
 , /* mUsePerTokenSfB */ 0
 , /* mUseTmaStore */ 1
 , /* mUseTwoTmaLoadWarps */ 1
 , /* mUseTwoMmaWarps */ 0
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
-, /* mSfLayoutB */ trtllm::gen::SfLayout(3)
-, /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfLayoutB */ trtllm::gen::SfLayout(1)
+, /* mSfLayoutC */ trtllm::gen::SfLayout(1)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -3588,73 +3639,79 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 1
-, /* mGridWaitForPrimaryRouting */ 1
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
+, /* mGridWaitForPrimaryRouting */ 0
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x16x128_EpiTile64x16_Mma64x16x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x16x128_EpiTile64x16_Mma64x16x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin_len, 82944, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x16x128_EpiTile64x16_Mma64x16x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin_len, 153600, "BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a", 480, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
 , /* mDtypeAcc */ trtllm::gen::Dtype(1056776)
-, /* mDtypeA */ trtllm::gen::Dtype(1050629)
-, /* mDtypeB */ trtllm::gen::Dtype(1050629)
-, /* mDtypeC */ trtllm::gen::Dtype(1050629)
+, /* mDtypeA */ trtllm::gen::Dtype(17826828)
+, /* mDtypeB */ trtllm::gen::Dtype(17827853)
+, /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(17826828)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(17827853)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
 , /* mEpilogueLdtmDps */ 16
 , /* mEpilogueLdtmBits */ 256
-, /* mEpilogueTileM */ 64
-, /* mEpilogueTileN */ 16
+, /* mEpilogueTileM */ 128
+, /* mEpilogueTileN */ 8
 , /* mGridTriggerSecondaryA */ 0
 , /* mGridTriggerSecondaryB */ 1
 , /* mGridWaitForPrimaryEarlyExit */ 1
 , /* mGridWaitForPrimaryA */ 0
 , /* mGridWaitForPrimaryB */ 1
 , /* mHoistLoadTaskInit */ 1
-, /* mHoistMmaTaskTryWaits */ 1
+, /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
-, /* mMmaKind */ trtllm::gen::MmaKind(2)
-, /* mMmaM */ 64
-, /* mMmaN */ 16
+, /* mMmaKind */ trtllm::gen::MmaKind(5)
+, /* mMmaM */ 128
+, /* mMmaN */ 8
 , /* mMockAllReduce */ 0
 , /* mN */ 256
 , /* mNumSlicesForSplitK */ 1
 , /* mNumSlicesForSliceK */ 1
 , /* mNumStages */ 4
-, /* mNumStagesMma */ 4
-, /* mNumStagesMmaWithinWorkTile */ 2
+, /* mNumStagesMma */ 2
+, /* mNumStagesMmaWithinWorkTile */ 1
 , /* mNumStagesMmaAcrossWorkTile */ 2
 , /* mNumStagesWorkId */ 3
 , /* mOutputDebugTensors */ 0
-, /* mUseShuffledMatrixA */ 0
+, /* mUseShuffledMatrixA */ 1
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
-, /* mTileN */ 16
-, /* mTileK */ 128
+, /* mTileN */ 8
+, /* mTileK */ 256
 , /* mUseUnrollLoop2xForMma */ 0
 , /* mUseCustomMmaSchedule */ 1
 , /* mUseHoistTryWaitForCustomMmaSchedule */ 0
-, /* mUseDeepSeekFp8 */ 1
+, /* mUseDeepSeekFp8 */ 0
 , /* mUsePerTokenSfA */ 0
 , /* mUsePerTokenSfB */ 0
 , /* mUseTmaStore */ 1
 , /* mUseTwoTmaLoadWarps */ 1
 , /* mUseTwoMmaWarps */ 0
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
-, /* mSfLayoutB */ trtllm::gen::SfLayout(3)
-, /* mSfLayoutC */ trtllm::gen::SfLayout(3)
-, /* mTileScheduler */ gemm::TileScheduler(0)
+, /* mSfLayoutB */ trtllm::gen::SfLayout(1)
+, /* mSfLayoutC */ trtllm::gen::SfLayout(1)
+, /* mSfReshapeFactor */ 1
+, /* mTileScheduler */ gemm::TileScheduler(1)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
 , /* mBatchedN */ {}
@@ -3662,72 +3719,78 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 1
-, /* mGridWaitForPrimaryRouting */ 1
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
+, /* mGridWaitForPrimaryRouting */ 0
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x16x128u2_EpiTile64x16_Mma64x16x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x16x128u2_EpiTile64x16_Mma64x16x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin_len, 82944, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x16x128u2_EpiTile64x16_Mma64x16x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x256u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x256u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin_len, 153600, "BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x256u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a", 448, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
 , /* mDtypeAcc */ trtllm::gen::Dtype(1056776)
-, /* mDtypeA */ trtllm::gen::Dtype(1050629)
-, /* mDtypeB */ trtllm::gen::Dtype(1050629)
-, /* mDtypeC */ trtllm::gen::Dtype(1050629)
+, /* mDtypeA */ trtllm::gen::Dtype(17826828)
+, /* mDtypeB */ trtllm::gen::Dtype(17827853)
+, /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(17826828)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(17827853)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
 , /* mEpilogueLdtmDps */ 16
 , /* mEpilogueLdtmBits */ 256
-, /* mEpilogueTileM */ 64
-, /* mEpilogueTileN */ 16
+, /* mEpilogueTileM */ 128
+, /* mEpilogueTileN */ 8
 , /* mGridTriggerSecondaryA */ 0
 , /* mGridTriggerSecondaryB */ 1
 , /* mGridWaitForPrimaryEarlyExit */ 1
 , /* mGridWaitForPrimaryA */ 0
 , /* mGridWaitForPrimaryB */ 1
 , /* mHoistLoadTaskInit */ 1
-, /* mHoistMmaTaskTryWaits */ 1
+, /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
-, /* mMmaKind */ trtllm::gen::MmaKind(2)
-, /* mMmaM */ 64
-, /* mMmaN */ 16
+, /* mMmaKind */ trtllm::gen::MmaKind(5)
+, /* mMmaM */ 128
+, /* mMmaN */ 8
 , /* mMockAllReduce */ 0
 , /* mN */ 256
 , /* mNumSlicesForSplitK */ 1
 , /* mNumSlicesForSliceK */ 1
 , /* mNumStages */ 4
-, /* mNumStagesMma */ 4
-, /* mNumStagesMmaWithinWorkTile */ 2
-, /* mNumStagesMmaAcrossWorkTile */ 2
+, /* mNumStagesMma */ 1
+, /* mNumStagesMmaWithinWorkTile */ 1
+, /* mNumStagesMmaAcrossWorkTile */ 1
 , /* mNumStagesWorkId */ 3
 , /* mOutputDebugTensors */ 0
-, /* mUseShuffledMatrixA */ 0
+, /* mUseShuffledMatrixA */ 1
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
-, /* mTileN */ 16
-, /* mTileK */ 128
+, /* mTileN */ 8
+, /* mTileK */ 256
 , /* mUseUnrollLoop2xForMma */ 1
 , /* mUseCustomMmaSchedule */ 1
 , /* mUseHoistTryWaitForCustomMmaSchedule */ 0
-, /* mUseDeepSeekFp8 */ 1
+, /* mUseDeepSeekFp8 */ 0
 , /* mUsePerTokenSfA */ 0
 , /* mUsePerTokenSfB */ 0
 , /* mUseTmaStore */ 1
 , /* mUseTwoTmaLoadWarps */ 1
 , /* mUseTwoMmaWarps */ 0
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
-, /* mSfLayoutB */ trtllm::gen::SfLayout(3)
-, /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfLayoutB */ trtllm::gen::SfLayout(1)
+, /* mSfLayoutC */ trtllm::gen::SfLayout(1)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -3736,161 +3799,1056 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 1
-, /* mGridWaitForPrimaryRouting */ 1
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
+, /* mGridWaitForPrimaryRouting */ 0
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x32x128_EpiTile64x32_Mma64x32x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x32x128_EpiTile64x32_Mma64x32x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin_len, 93184, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x32x128_EpiTile64x32_Mma64x32x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x256u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x256u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin_len, 153600, "BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x256u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a", 480, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
 , /* mDtypeAcc */ trtllm::gen::Dtype(1056776)
-, /* mDtypeA */ trtllm::gen::Dtype(1050629)
-, /* mDtypeB */ trtllm::gen::Dtype(1050629)
-, /* mDtypeC */ trtllm::gen::Dtype(1050629)
+, /* mDtypeA */ trtllm::gen::Dtype(17826828)
+, /* mDtypeB */ trtllm::gen::Dtype(17827853)
+, /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(17826828)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(17827853)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
 , /* mEpilogueLdtmDps */ 16
 , /* mEpilogueLdtmBits */ 256
-, /* mEpilogueTileM */ 64
-, /* mEpilogueTileN */ 32
+, /* mEpilogueTileM */ 128
+, /* mEpilogueTileN */ 8
 , /* mGridTriggerSecondaryA */ 0
 , /* mGridTriggerSecondaryB */ 1
 , /* mGridWaitForPrimaryEarlyExit */ 1
 , /* mGridWaitForPrimaryA */ 0
 , /* mGridWaitForPrimaryB */ 1
 , /* mHoistLoadTaskInit */ 1
-, /* mHoistMmaTaskTryWaits */ 1
+, /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
-, /* mMmaKind */ trtllm::gen::MmaKind(2)
-, /* mMmaM */ 64
-, /* mMmaN */ 32
+, /* mMmaKind */ trtllm::gen::MmaKind(5)
+, /* mMmaM */ 128
+, /* mMmaN */ 8
 , /* mMockAllReduce */ 0
 , /* mN */ 256
 , /* mNumSlicesForSplitK */ 1
 , /* mNumSlicesForSliceK */ 1
 , /* mNumStages */ 4
-, /* mNumStagesMma */ 4
-, /* mNumStagesMmaWithinWorkTile */ 2
+, /* mNumStagesMma */ 2
+, /* mNumStagesMmaWithinWorkTile */ 1
+, /* mNumStagesMmaAcrossWorkTile */ 2
+, /* mNumStagesWorkId */ 3
+, /* mOutputDebugTensors */ 0
+, /* mUseShuffledMatrixA */ 1
+, /* mSliceK */ 0
+, /* mSplitK */ gemm::SplitK(0)
+, /* mTransposeMmaOutput */ 1
+, /* mTileM */ 128
+, /* mTileN */ 8
+, /* mTileK */ 256
+, /* mUseUnrollLoop2xForMma */ 1
+, /* mUseCustomMmaSchedule */ 1
+, /* mUseHoistTryWaitForCustomMmaSchedule */ 0
+, /* mUseDeepSeekFp8 */ 0
+, /* mUsePerTokenSfA */ 0
+, /* mUsePerTokenSfB */ 0
+, /* mUseTmaStore */ 1
+, /* mUseTwoTmaLoadWarps */ 1
+, /* mUseTwoMmaWarps */ 0
+, /* mSfLayoutA */ trtllm::gen::SfLayout(3)
+, /* mSfLayoutB */ trtllm::gen::SfLayout(1)
+, /* mSfLayoutC */ trtllm::gen::SfLayout(1)
+, /* mSfReshapeFactor */ 1
+, /* mTileScheduler */ gemm::TileScheduler(1)
+, /* mActType */ gemmGatedAct::ActType(0)
+, /* mBatchedM */ {}
+, /* mBatchedN */ {}
+, /* mBatchMode */ batchedGemm::BatchedGemmOptions::BatchMode(1)
+, /* mNumBatches */ 2
+, /* mIsStaticBatch */ 0
+, /* mNumTokens */ 2
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
+, /* mGridWaitForPrimaryRouting */ 0
+, /* mFusedAct */ 0
+, /* mNumRegsPerThreadNonEpilogueWarp */ 0
+, /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
+ }, gemm::SmVersion::Sm100a },
+{BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin_len, 225280, "BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a", 448, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
+, /* mClusterDimX */ 1
+, /* mClusterDimY */ 1
+, /* mClusterDimZ */ 1
+, /* mDtypeAcc */ trtllm::gen::Dtype(1056776)
+, /* mDtypeA */ trtllm::gen::Dtype(17826828)
+, /* mDtypeB */ trtllm::gen::Dtype(17827853)
+, /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(17826828)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(17827853)
+, /* mEnablesEarlyExit */ 1
+, /* mEnablesDelayedEarlyExit */ 0
+, /* mEnablesGlobalPtxKnobs */ 1
+, /* mEpilogueLdtmDps */ 16
+, /* mEpilogueLdtmBits */ 256
+, /* mEpilogueTileM */ 128
+, /* mEpilogueTileN */ 8
+, /* mGridTriggerSecondaryA */ 0
+, /* mGridTriggerSecondaryB */ 1
+, /* mGridWaitForPrimaryEarlyExit */ 1
+, /* mGridWaitForPrimaryA */ 0
+, /* mGridWaitForPrimaryB */ 1
+, /* mHoistLoadTaskInit */ 1
+, /* mHoistMmaTaskTryWaits */ 0
+, /* mK */ 2048
+, /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
+, /* mM */ 256
+, /* mMmaK */ 32
+, /* mMmaKind */ trtllm::gen::MmaKind(5)
+, /* mMmaM */ 128
+, /* mMmaN */ 8
+, /* mMockAllReduce */ 0
+, /* mN */ 256
+, /* mNumSlicesForSplitK */ 1
+, /* mNumSlicesForSliceK */ 1
+, /* mNumStages */ 3
+, /* mNumStagesMma */ 1
+, /* mNumStagesMmaWithinWorkTile */ 1
+, /* mNumStagesMmaAcrossWorkTile */ 1
+, /* mNumStagesWorkId */ 3
+, /* mOutputDebugTensors */ 0
+, /* mUseShuffledMatrixA */ 1
+, /* mSliceK */ 0
+, /* mSplitK */ gemm::SplitK(0)
+, /* mTransposeMmaOutput */ 1
+, /* mTileM */ 128
+, /* mTileN */ 8
+, /* mTileK */ 512
+, /* mUseUnrollLoop2xForMma */ 0
+, /* mUseCustomMmaSchedule */ 1
+, /* mUseHoistTryWaitForCustomMmaSchedule */ 0
+, /* mUseDeepSeekFp8 */ 0
+, /* mUsePerTokenSfA */ 0
+, /* mUsePerTokenSfB */ 0
+, /* mUseTmaStore */ 1
+, /* mUseTwoTmaLoadWarps */ 1
+, /* mUseTwoMmaWarps */ 0
+, /* mSfLayoutA */ trtllm::gen::SfLayout(3)
+, /* mSfLayoutB */ trtllm::gen::SfLayout(1)
+, /* mSfLayoutC */ trtllm::gen::SfLayout(1)
+, /* mSfReshapeFactor */ 1
+, /* mTileScheduler */ gemm::TileScheduler(0)
+, /* mActType */ gemmGatedAct::ActType(0)
+, /* mBatchedM */ {}
+, /* mBatchedN */ {}
+, /* mBatchMode */ batchedGemm::BatchedGemmOptions::BatchMode(1)
+, /* mNumBatches */ 2
+, /* mIsStaticBatch */ 0
+, /* mNumTokens */ 2
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
+, /* mGridWaitForPrimaryRouting */ 0
+, /* mFusedAct */ 0
+, /* mNumRegsPerThreadNonEpilogueWarp */ 0
+, /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
+ }, gemm::SmVersion::Sm100a },
+{BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin_len, 225280, "BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a", 480, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
+, /* mClusterDimX */ 1
+, /* mClusterDimY */ 1
+, /* mClusterDimZ */ 1
+, /* mDtypeAcc */ trtllm::gen::Dtype(1056776)
+, /* mDtypeA */ trtllm::gen::Dtype(17826828)
+, /* mDtypeB */ trtllm::gen::Dtype(17827853)
+, /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(17826828)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(17827853)
+, /* mEnablesEarlyExit */ 1
+, /* mEnablesDelayedEarlyExit */ 0
+, /* mEnablesGlobalPtxKnobs */ 1
+, /* mEpilogueLdtmDps */ 16
+, /* mEpilogueLdtmBits */ 256
+, /* mEpilogueTileM */ 128
+, /* mEpilogueTileN */ 8
+, /* mGridTriggerSecondaryA */ 0
+, /* mGridTriggerSecondaryB */ 1
+, /* mGridWaitForPrimaryEarlyExit */ 1
+, /* mGridWaitForPrimaryA */ 0
+, /* mGridWaitForPrimaryB */ 1
+, /* mHoistLoadTaskInit */ 1
+, /* mHoistMmaTaskTryWaits */ 0
+, /* mK */ 2048
+, /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
+, /* mM */ 256
+, /* mMmaK */ 32
+, /* mMmaKind */ trtllm::gen::MmaKind(5)
+, /* mMmaM */ 128
+, /* mMmaN */ 8
+, /* mMockAllReduce */ 0
+, /* mN */ 256
+, /* mNumSlicesForSplitK */ 1
+, /* mNumSlicesForSliceK */ 1
+, /* mNumStages */ 3
+, /* mNumStagesMma */ 2
+, /* mNumStagesMmaWithinWorkTile */ 1
 , /* mNumStagesMmaAcrossWorkTile */ 2
 , /* mNumStagesWorkId */ 3
 , /* mOutputDebugTensors */ 0
-, /* mUseShuffledMatrixA */ 0
+, /* mUseShuffledMatrixA */ 1
+, /* mSliceK */ 0
+, /* mSplitK */ gemm::SplitK(0)
+, /* mTransposeMmaOutput */ 1
+, /* mTileM */ 128
+, /* mTileN */ 8
+, /* mTileK */ 512
+, /* mUseUnrollLoop2xForMma */ 0
+, /* mUseCustomMmaSchedule */ 1
+, /* mUseHoistTryWaitForCustomMmaSchedule */ 0
+, /* mUseDeepSeekFp8 */ 0
+, /* mUsePerTokenSfA */ 0
+, /* mUsePerTokenSfB */ 0
+, /* mUseTmaStore */ 1
+, /* mUseTwoTmaLoadWarps */ 1
+, /* mUseTwoMmaWarps */ 0
+, /* mSfLayoutA */ trtllm::gen::SfLayout(3)
+, /* mSfLayoutB */ trtllm::gen::SfLayout(1)
+, /* mSfLayoutC */ trtllm::gen::SfLayout(1)
+, /* mSfReshapeFactor */ 1
+, /* mTileScheduler */ gemm::TileScheduler(1)
+, /* mActType */ gemmGatedAct::ActType(0)
+, /* mBatchedM */ {}
+, /* mBatchedN */ {}
+, /* mBatchMode */ batchedGemm::BatchedGemmOptions::BatchMode(1)
+, /* mNumBatches */ 2
+, /* mIsStaticBatch */ 0
+, /* mNumTokens */ 2
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
+, /* mGridWaitForPrimaryRouting */ 0
+, /* mFusedAct */ 0
+, /* mNumRegsPerThreadNonEpilogueWarp */ 0
+, /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
+ }, gemm::SmVersion::Sm100a },
+{BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin_len, 225280, "BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a", 448, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
+, /* mClusterDimX */ 1
+, /* mClusterDimY */ 1
+, /* mClusterDimZ */ 1
+, /* mDtypeAcc */ trtllm::gen::Dtype(1056776)
+, /* mDtypeA */ trtllm::gen::Dtype(17826828)
+, /* mDtypeB */ trtllm::gen::Dtype(17827853)
+, /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(17826828)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(17827853)
+, /* mEnablesEarlyExit */ 1
+, /* mEnablesDelayedEarlyExit */ 0
+, /* mEnablesGlobalPtxKnobs */ 1
+, /* mEpilogueLdtmDps */ 16
+, /* mEpilogueLdtmBits */ 256
+, /* mEpilogueTileM */ 128
+, /* mEpilogueTileN */ 8
+, /* mGridTriggerSecondaryA */ 0
+, /* mGridTriggerSecondaryB */ 1
+, /* mGridWaitForPrimaryEarlyExit */ 1
+, /* mGridWaitForPrimaryA */ 0
+, /* mGridWaitForPrimaryB */ 1
+, /* mHoistLoadTaskInit */ 1
+, /* mHoistMmaTaskTryWaits */ 0
+, /* mK */ 2048
+, /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
+, /* mM */ 256
+, /* mMmaK */ 32
+, /* mMmaKind */ trtllm::gen::MmaKind(5)
+, /* mMmaM */ 128
+, /* mMmaN */ 8
+, /* mMockAllReduce */ 0
+, /* mN */ 256
+, /* mNumSlicesForSplitK */ 1
+, /* mNumSlicesForSliceK */ 1
+, /* mNumStages */ 3
+, /* mNumStagesMma */ 1
+, /* mNumStagesMmaWithinWorkTile */ 1
+, /* mNumStagesMmaAcrossWorkTile */ 1
+, /* mNumStagesWorkId */ 3
+, /* mOutputDebugTensors */ 0
+, /* mUseShuffledMatrixA */ 1
+, /* mSliceK */ 0
+, /* mSplitK */ gemm::SplitK(0)
+, /* mTransposeMmaOutput */ 1
+, /* mTileM */ 128
+, /* mTileN */ 8
+, /* mTileK */ 512
+, /* mUseUnrollLoop2xForMma */ 1
+, /* mUseCustomMmaSchedule */ 1
+, /* mUseHoistTryWaitForCustomMmaSchedule */ 0
+, /* mUseDeepSeekFp8 */ 0
+, /* mUsePerTokenSfA */ 0
+, /* mUsePerTokenSfB */ 0
+, /* mUseTmaStore */ 1
+, /* mUseTwoTmaLoadWarps */ 1
+, /* mUseTwoMmaWarps */ 0
+, /* mSfLayoutA */ trtllm::gen::SfLayout(3)
+, /* mSfLayoutB */ trtllm::gen::SfLayout(1)
+, /* mSfLayoutC */ trtllm::gen::SfLayout(1)
+, /* mSfReshapeFactor */ 1
+, /* mTileScheduler */ gemm::TileScheduler(0)
+, /* mActType */ gemmGatedAct::ActType(0)
+, /* mBatchedM */ {}
+, /* mBatchedN */ {}
+, /* mBatchMode */ batchedGemm::BatchedGemmOptions::BatchMode(1)
+, /* mNumBatches */ 2
+, /* mIsStaticBatch */ 0
+, /* mNumTokens */ 2
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
+, /* mGridWaitForPrimaryRouting */ 0
+, /* mFusedAct */ 0
+, /* mNumRegsPerThreadNonEpilogueWarp */ 0
+, /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
+ }, gemm::SmVersion::Sm100a },
+{BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin, BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin_len, 225280, "BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a", 480, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
+, /* mClusterDimX */ 1
+, /* mClusterDimY */ 1
+, /* mClusterDimZ */ 1
+, /* mDtypeAcc */ trtllm::gen::Dtype(1056776)
+, /* mDtypeA */ trtllm::gen::Dtype(17826828)
+, /* mDtypeB */ trtllm::gen::Dtype(17827853)
+, /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(17826828)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(17827853)
+, /* mEnablesEarlyExit */ 1
+, /* mEnablesDelayedEarlyExit */ 0
+, /* mEnablesGlobalPtxKnobs */ 1
+, /* mEpilogueLdtmDps */ 16
+, /* mEpilogueLdtmBits */ 256
+, /* mEpilogueTileM */ 128
+, /* mEpilogueTileN */ 8
+, /* mGridTriggerSecondaryA */ 0
+, /* mGridTriggerSecondaryB */ 1
+, /* mGridWaitForPrimaryEarlyExit */ 1
+, /* mGridWaitForPrimaryA */ 0
+, /* mGridWaitForPrimaryB */ 1
+, /* mHoistLoadTaskInit */ 1
+, /* mHoistMmaTaskTryWaits */ 0
+, /* mK */ 2048
+, /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
+, /* mM */ 256
+, /* mMmaK */ 32
+, /* mMmaKind */ trtllm::gen::MmaKind(5)
+, /* mMmaM */ 128
+, /* mMmaN */ 8
+, /* mMockAllReduce */ 0
+, /* mN */ 256
+, /* mNumSlicesForSplitK */ 1
+, /* mNumSlicesForSliceK */ 1
+, /* mNumStages */ 3
+, /* mNumStagesMma */ 2
+, /* mNumStagesMmaWithinWorkTile */ 1
+, /* mNumStagesMmaAcrossWorkTile */ 2
+, /* mNumStagesWorkId */ 3
+, /* mOutputDebugTensors */ 0
+, /* mUseShuffledMatrixA */ 1
+, /* mSliceK */ 0
+, /* mSplitK */ gemm::SplitK(0)
+, /* mTransposeMmaOutput */ 1
+, /* mTileM */ 128
+, /* mTileN */ 8
+, /* mTileK */ 512
+, /* mUseUnrollLoop2xForMma */ 1
+, /* mUseCustomMmaSchedule */ 1
+, /* mUseHoistTryWaitForCustomMmaSchedule */ 0
+, /* mUseDeepSeekFp8 */ 0
+, /* mUsePerTokenSfA */ 0
+, /* mUsePerTokenSfB */ 0
+, /* mUseTmaStore */ 1
+, /* mUseTwoTmaLoadWarps */ 1
+, /* mUseTwoMmaWarps */ 0
+, /* mSfLayoutA */ trtllm::gen::SfLayout(3)
+, /* mSfLayoutB */ trtllm::gen::SfLayout(1)
+, /* mSfLayoutC */ trtllm::gen::SfLayout(1)
+, /* mSfReshapeFactor */ 1
+, /* mTileScheduler */ gemm::TileScheduler(1)
+, /* mActType */ gemmGatedAct::ActType(0)
+, /* mBatchedM */ {}
+, /* mBatchedN */ {}
+, /* mBatchMode */ batchedGemm::BatchedGemmOptions::BatchMode(1)
+, /* mNumBatches */ 2
+, /* mIsStaticBatch */ 0
+, /* mNumTokens */ 2
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
+, /* mGridWaitForPrimaryRouting */ 0
+, /* mFusedAct */ 0
+, /* mNumRegsPerThreadNonEpilogueWarp */ 0
+, /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
+ }, gemm::SmVersion::Sm100a },
+{BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin, BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin_len, 165888, "BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a", 480, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
+, /* mClusterDimX */ 1
+, /* mClusterDimY */ 1
+, /* mClusterDimZ */ 1
+, /* mDtypeAcc */ trtllm::gen::Dtype(1056776)
+, /* mDtypeA */ trtllm::gen::Dtype(17826818)
+, /* mDtypeB */ trtllm::gen::Dtype(17826818)
+, /* mDtypeC */ trtllm::gen::Dtype(17826818)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(17826818)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(17826818)
+, /* mEnablesEarlyExit */ 1
+, /* mEnablesDelayedEarlyExit */ 0
+, /* mEnablesGlobalPtxKnobs */ 1
+, /* mEpilogueLdtmDps */ 16
+, /* mEpilogueLdtmBits */ 256
+, /* mEpilogueTileM */ 128
+, /* mEpilogueTileN */ 8
+, /* mGridTriggerSecondaryA */ 0
+, /* mGridTriggerSecondaryB */ 1
+, /* mGridWaitForPrimaryEarlyExit */ 1
+, /* mGridWaitForPrimaryA */ 0
+, /* mGridWaitForPrimaryB */ 1
+, /* mHoistLoadTaskInit */ 1
+, /* mHoistMmaTaskTryWaits */ 0
+, /* mK */ 2048
+, /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
+, /* mM */ 256
+, /* mMmaK */ 64
+, /* mMmaKind */ trtllm::gen::MmaKind(4)
+, /* mMmaM */ 128
+, /* mMmaN */ 8
+, /* mMockAllReduce */ 0
+, /* mN */ 256
+, /* mNumSlicesForSplitK */ 1
+, /* mNumSlicesForSliceK */ 1
+, /* mNumStages */ 4
+, /* mNumStagesMma */ 1
+, /* mNumStagesMmaWithinWorkTile */ 1
+, /* mNumStagesMmaAcrossWorkTile */ 1
+, /* mNumStagesWorkId */ 3
+, /* mOutputDebugTensors */ 0
+, /* mUseShuffledMatrixA */ 1
+, /* mSliceK */ 0
+, /* mSplitK */ gemm::SplitK(0)
+, /* mTransposeMmaOutput */ 1
+, /* mTileM */ 128
+, /* mTileN */ 8
+, /* mTileK */ 512
+, /* mUseUnrollLoop2xForMma */ 0
+, /* mUseCustomMmaSchedule */ 1
+, /* mUseHoistTryWaitForCustomMmaSchedule */ 0
+, /* mUseDeepSeekFp8 */ 0
+, /* mUsePerTokenSfA */ 0
+, /* mUsePerTokenSfB */ 0
+, /* mUseTmaStore */ 1
+, /* mUseTwoTmaLoadWarps */ 1
+, /* mUseTwoMmaWarps */ 0
+, /* mSfLayoutA */ trtllm::gen::SfLayout(3)
+, /* mSfLayoutB */ trtllm::gen::SfLayout(0)
+, /* mSfLayoutC */ trtllm::gen::SfLayout(1)
+, /* mSfReshapeFactor */ 1
+, /* mTileScheduler */ gemm::TileScheduler(0)
+, /* mActType */ gemmGatedAct::ActType(0)
+, /* mBatchedM */ {}
+, /* mBatchedN */ {}
+, /* mBatchMode */ batchedGemm::BatchedGemmOptions::BatchMode(1)
+, /* mNumBatches */ 2
+, /* mIsStaticBatch */ 0
+, /* mNumTokens */ 2
+, /* mRouteImpl */ batchedGemm::RouteImpl(1)
+, /* mGridWaitForPrimaryRouting */ 1
+, /* mFusedAct */ 1
+, /* mNumRegsPerThreadNonEpilogueWarp */ 0
+, /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
+ }, gemm::SmVersion::Sm100a },
+{BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin, BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin_len, 165888, "BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a", 448, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
+, /* mClusterDimX */ 1
+, /* mClusterDimY */ 1
+, /* mClusterDimZ */ 1
+, /* mDtypeAcc */ trtllm::gen::Dtype(1056776)
+, /* mDtypeA */ trtllm::gen::Dtype(17826818)
+, /* mDtypeB */ trtllm::gen::Dtype(17826818)
+, /* mDtypeC */ trtllm::gen::Dtype(17826818)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(17826818)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(17826818)
+, /* mEnablesEarlyExit */ 0
+, /* mEnablesDelayedEarlyExit */ 0
+, /* mEnablesGlobalPtxKnobs */ 1
+, /* mEpilogueLdtmDps */ 16
+, /* mEpilogueLdtmBits */ 256
+, /* mEpilogueTileM */ 128
+, /* mEpilogueTileN */ 8
+, /* mGridTriggerSecondaryA */ 0
+, /* mGridTriggerSecondaryB */ 1
+, /* mGridWaitForPrimaryEarlyExit */ 1
+, /* mGridWaitForPrimaryA */ 0
+, /* mGridWaitForPrimaryB */ 1
+, /* mHoistLoadTaskInit */ 1
+, /* mHoistMmaTaskTryWaits */ 0
+, /* mK */ 2048
+, /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
+, /* mM */ 256
+, /* mMmaK */ 64
+, /* mMmaKind */ trtllm::gen::MmaKind(4)
+, /* mMmaM */ 128
+, /* mMmaN */ 8
+, /* mMockAllReduce */ 0
+, /* mN */ 256
+, /* mNumSlicesForSplitK */ 1
+, /* mNumSlicesForSliceK */ 1
+, /* mNumStages */ 4
+, /* mNumStagesMma */ 1
+, /* mNumStagesMmaWithinWorkTile */ 1
+, /* mNumStagesMmaAcrossWorkTile */ 1
+, /* mNumStagesWorkId */ 3
+, /* mOutputDebugTensors */ 0
+, /* mUseShuffledMatrixA */ 1
+, /* mSliceK */ 0
+, /* mSplitK */ gemm::SplitK(0)
+, /* mTransposeMmaOutput */ 1
+, /* mTileM */ 128
+, /* mTileN */ 8
+, /* mTileK */ 512
+, /* mUseUnrollLoop2xForMma */ 0
+, /* mUseCustomMmaSchedule */ 1
+, /* mUseHoistTryWaitForCustomMmaSchedule */ 0
+, /* mUseDeepSeekFp8 */ 0
+, /* mUsePerTokenSfA */ 0
+, /* mUsePerTokenSfB */ 0
+, /* mUseTmaStore */ 1
+, /* mUseTwoTmaLoadWarps */ 1
+, /* mUseTwoMmaWarps */ 0
+, /* mSfLayoutA */ trtllm::gen::SfLayout(3)
+, /* mSfLayoutB */ trtllm::gen::SfLayout(1)
+, /* mSfLayoutC */ trtllm::gen::SfLayout(1)
+, /* mSfReshapeFactor */ 1
+, /* mTileScheduler */ gemm::TileScheduler(0)
+, /* mActType */ gemmGatedAct::ActType(0)
+, /* mBatchedM */ {}
+, /* mBatchedN */ {}
+, /* mBatchMode */ batchedGemm::BatchedGemmOptions::BatchMode(1)
+, /* mNumBatches */ 2
+, /* mIsStaticBatch */ 1
+, /* mNumTokens */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
+, /* mGridWaitForPrimaryRouting */ 1
+, /* mFusedAct */ 0
+, /* mNumRegsPerThreadNonEpilogueWarp */ 0
+, /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
+ }, gemm::SmVersion::Sm100a },
+{BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin, BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin_len, 165888, "BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a", 480, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
+, /* mClusterDimX */ 1
+, /* mClusterDimY */ 1
+, /* mClusterDimZ */ 1
+, /* mDtypeAcc */ trtllm::gen::Dtype(1056776)
+, /* mDtypeA */ trtllm::gen::Dtype(17826818)
+, /* mDtypeB */ trtllm::gen::Dtype(17826818)
+, /* mDtypeC */ trtllm::gen::Dtype(17826818)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(17826818)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(17826818)
+, /* mEnablesEarlyExit */ 1
+, /* mEnablesDelayedEarlyExit */ 0
+, /* mEnablesGlobalPtxKnobs */ 1
+, /* mEpilogueLdtmDps */ 16
+, /* mEpilogueLdtmBits */ 256
+, /* mEpilogueTileM */ 128
+, /* mEpilogueTileN */ 8
+, /* mGridTriggerSecondaryA */ 0
+, /* mGridTriggerSecondaryB */ 1
+, /* mGridWaitForPrimaryEarlyExit */ 1
+, /* mGridWaitForPrimaryA */ 0
+, /* mGridWaitForPrimaryB */ 1
+, /* mHoistLoadTaskInit */ 1
+, /* mHoistMmaTaskTryWaits */ 0
+, /* mK */ 2048
+, /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
+, /* mM */ 256
+, /* mMmaK */ 64
+, /* mMmaKind */ trtllm::gen::MmaKind(4)
+, /* mMmaM */ 128
+, /* mMmaN */ 8
+, /* mMockAllReduce */ 0
+, /* mN */ 256
+, /* mNumSlicesForSplitK */ 1
+, /* mNumSlicesForSliceK */ 1
+, /* mNumStages */ 4
+, /* mNumStagesMma */ 1
+, /* mNumStagesMmaWithinWorkTile */ 1
+, /* mNumStagesMmaAcrossWorkTile */ 1
+, /* mNumStagesWorkId */ 3
+, /* mOutputDebugTensors */ 0
+, /* mUseShuffledMatrixA */ 1
+, /* mSliceK */ 0
+, /* mSplitK */ gemm::SplitK(0)
+, /* mTransposeMmaOutput */ 1
+, /* mTileM */ 128
+, /* mTileN */ 8
+, /* mTileK */ 512
+, /* mUseUnrollLoop2xForMma */ 1
+, /* mUseCustomMmaSchedule */ 1
+, /* mUseHoistTryWaitForCustomMmaSchedule */ 0
+, /* mUseDeepSeekFp8 */ 0
+, /* mUsePerTokenSfA */ 0
+, /* mUsePerTokenSfB */ 0
+, /* mUseTmaStore */ 1
+, /* mUseTwoTmaLoadWarps */ 1
+, /* mUseTwoMmaWarps */ 0
+, /* mSfLayoutA */ trtllm::gen::SfLayout(3)
+, /* mSfLayoutB */ trtllm::gen::SfLayout(0)
+, /* mSfLayoutC */ trtllm::gen::SfLayout(1)
+, /* mSfReshapeFactor */ 1
+, /* mTileScheduler */ gemm::TileScheduler(0)
+, /* mActType */ gemmGatedAct::ActType(0)
+, /* mBatchedM */ {}
+, /* mBatchedN */ {}
+, /* mBatchMode */ batchedGemm::BatchedGemmOptions::BatchMode(1)
+, /* mNumBatches */ 2
+, /* mIsStaticBatch */ 0
+, /* mNumTokens */ 2
+, /* mRouteImpl */ batchedGemm::RouteImpl(1)
+, /* mGridWaitForPrimaryRouting */ 1
+, /* mFusedAct */ 1
+, /* mNumRegsPerThreadNonEpilogueWarp */ 0
+, /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
+ }, gemm::SmVersion::Sm100a },
+{BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin, BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin_len, 165888, "BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a", 480, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
+, /* mClusterDimX */ 1
+, /* mClusterDimY */ 1
+, /* mClusterDimZ */ 1
+, /* mDtypeAcc */ trtllm::gen::Dtype(1056776)
+, /* mDtypeA */ trtllm::gen::Dtype(17826818)
+, /* mDtypeB */ trtllm::gen::Dtype(17826818)
+, /* mDtypeC */ trtllm::gen::Dtype(17826818)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(17826818)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(17826818)
+, /* mEnablesEarlyExit */ 1
+, /* mEnablesDelayedEarlyExit */ 0
+, /* mEnablesGlobalPtxKnobs */ 1
+, /* mEpilogueLdtmDps */ 16
+, /* mEpilogueLdtmBits */ 256
+, /* mEpilogueTileM */ 128
+, /* mEpilogueTileN */ 8
+, /* mGridTriggerSecondaryA */ 0
+, /* mGridTriggerSecondaryB */ 1
+, /* mGridWaitForPrimaryEarlyExit */ 1
+, /* mGridWaitForPrimaryA */ 0
+, /* mGridWaitForPrimaryB */ 1
+, /* mHoistLoadTaskInit */ 1
+, /* mHoistMmaTaskTryWaits */ 0
+, /* mK */ 2048
+, /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
+, /* mM */ 256
+, /* mMmaK */ 64
+, /* mMmaKind */ trtllm::gen::MmaKind(4)
+, /* mMmaM */ 128
+, /* mMmaN */ 8
+, /* mMockAllReduce */ 0
+, /* mN */ 256
+, /* mNumSlicesForSplitK */ 1
+, /* mNumSlicesForSliceK */ 1
+, /* mNumStages */ 4
+, /* mNumStagesMma */ 1
+, /* mNumStagesMmaWithinWorkTile */ 1
+, /* mNumStagesMmaAcrossWorkTile */ 1
+, /* mNumStagesWorkId */ 3
+, /* mOutputDebugTensors */ 0
+, /* mUseShuffledMatrixA */ 1
+, /* mSliceK */ 0
+, /* mSplitK */ gemm::SplitK(0)
+, /* mTransposeMmaOutput */ 1
+, /* mTileM */ 128
+, /* mTileN */ 8
+, /* mTileK */ 512
+, /* mUseUnrollLoop2xForMma */ 1
+, /* mUseCustomMmaSchedule */ 1
+, /* mUseHoistTryWaitForCustomMmaSchedule */ 0
+, /* mUseDeepSeekFp8 */ 0
+, /* mUsePerTokenSfA */ 0
+, /* mUsePerTokenSfB */ 0
+, /* mUseTmaStore */ 1
+, /* mUseTwoTmaLoadWarps */ 1
+, /* mUseTwoMmaWarps */ 0
+, /* mSfLayoutA */ trtllm::gen::SfLayout(3)
+, /* mSfLayoutB */ trtllm::gen::SfLayout(0)
+, /* mSfLayoutC */ trtllm::gen::SfLayout(1)
+, /* mSfReshapeFactor */ 1
+, /* mTileScheduler */ gemm::TileScheduler(0)
+, /* mActType */ gemmGatedAct::ActType(0)
+, /* mBatchedM */ {}
+, /* mBatchedN */ {}
+, /* mBatchMode */ batchedGemm::BatchedGemmOptions::BatchMode(1)
+, /* mNumBatches */ 128
+, /* mIsStaticBatch */ 0
+, /* mNumTokens */ 32
+, /* mRouteImpl */ batchedGemm::RouteImpl(1)
+, /* mGridWaitForPrimaryRouting */ 1
+, /* mFusedAct */ 1
+, /* mNumRegsPerThreadNonEpilogueWarp */ 0
+, /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
+ }, gemm::SmVersion::Sm100a },
+{BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin, BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin_len, 165888, "BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a", 480, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
+, /* mClusterDimX */ 1
+, /* mClusterDimY */ 1
+, /* mClusterDimZ */ 1
+, /* mDtypeAcc */ trtllm::gen::Dtype(1056776)
+, /* mDtypeA */ trtllm::gen::Dtype(17826818)
+, /* mDtypeB */ trtllm::gen::Dtype(17826818)
+, /* mDtypeC */ trtllm::gen::Dtype(17826818)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(17826818)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(17826818)
+, /* mEnablesEarlyExit */ 1
+, /* mEnablesDelayedEarlyExit */ 0
+, /* mEnablesGlobalPtxKnobs */ 1
+, /* mEpilogueLdtmDps */ 16
+, /* mEpilogueLdtmBits */ 256
+, /* mEpilogueTileM */ 128
+, /* mEpilogueTileN */ 8
+, /* mGridTriggerSecondaryA */ 0
+, /* mGridTriggerSecondaryB */ 1
+, /* mGridWaitForPrimaryEarlyExit */ 1
+, /* mGridWaitForPrimaryA */ 0
+, /* mGridWaitForPrimaryB */ 1
+, /* mHoistLoadTaskInit */ 1
+, /* mHoistMmaTaskTryWaits */ 0
+, /* mK */ 2048
+, /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
+, /* mM */ 256
+, /* mMmaK */ 64
+, /* mMmaKind */ trtllm::gen::MmaKind(4)
+, /* mMmaM */ 128
+, /* mMmaN */ 8
+, /* mMockAllReduce */ 0
+, /* mN */ 256
+, /* mNumSlicesForSplitK */ 1
+, /* mNumSlicesForSliceK */ 1
+, /* mNumStages */ 4
+, /* mNumStagesMma */ 1
+, /* mNumStagesMmaWithinWorkTile */ 1
+, /* mNumStagesMmaAcrossWorkTile */ 1
+, /* mNumStagesWorkId */ 3
+, /* mOutputDebugTensors */ 0
+, /* mUseShuffledMatrixA */ 1
+, /* mSliceK */ 0
+, /* mSplitK */ gemm::SplitK(0)
+, /* mTransposeMmaOutput */ 1
+, /* mTileM */ 128
+, /* mTileN */ 8
+, /* mTileK */ 512
+, /* mUseUnrollLoop2xForMma */ 1
+, /* mUseCustomMmaSchedule */ 1
+, /* mUseHoistTryWaitForCustomMmaSchedule */ 0
+, /* mUseDeepSeekFp8 */ 0
+, /* mUsePerTokenSfA */ 0
+, /* mUsePerTokenSfB */ 0
+, /* mUseTmaStore */ 1
+, /* mUseTwoTmaLoadWarps */ 1
+, /* mUseTwoMmaWarps */ 0
+, /* mSfLayoutA */ trtllm::gen::SfLayout(3)
+, /* mSfLayoutB */ trtllm::gen::SfLayout(0)
+, /* mSfLayoutC */ trtllm::gen::SfLayout(1)
+, /* mSfReshapeFactor */ 1
+, /* mTileScheduler */ gemm::TileScheduler(0)
+, /* mActType */ gemmGatedAct::ActType(0)
+, /* mBatchedM */ {}
+, /* mBatchedN */ {}
+, /* mBatchMode */ batchedGemm::BatchedGemmOptions::BatchMode(1)
+, /* mNumBatches */ 128
+, /* mIsStaticBatch */ 0
+, /* mNumTokens */ 32
+, /* mRouteImpl */ batchedGemm::RouteImpl(1)
+, /* mGridWaitForPrimaryRouting */ 1
+, /* mFusedAct */ 1
+, /* mNumRegsPerThreadNonEpilogueWarp */ 0
+, /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
+ }, gemm::SmVersion::Sm100a },
+{BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin, BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin_len, 165888, "BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a", 480, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
+, /* mClusterDimX */ 1
+, /* mClusterDimY */ 1
+, /* mClusterDimZ */ 1
+, /* mDtypeAcc */ trtllm::gen::Dtype(1056776)
+, /* mDtypeA */ trtllm::gen::Dtype(17826818)
+, /* mDtypeB */ trtllm::gen::Dtype(17826818)
+, /* mDtypeC */ trtllm::gen::Dtype(17826818)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(17826818)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(17826818)
+, /* mEnablesEarlyExit */ 1
+, /* mEnablesDelayedEarlyExit */ 0
+, /* mEnablesGlobalPtxKnobs */ 1
+, /* mEpilogueLdtmDps */ 16
+, /* mEpilogueLdtmBits */ 256
+, /* mEpilogueTileM */ 128
+, /* mEpilogueTileN */ 8
+, /* mGridTriggerSecondaryA */ 0
+, /* mGridTriggerSecondaryB */ 1
+, /* mGridWaitForPrimaryEarlyExit */ 1
+, /* mGridWaitForPrimaryA */ 0
+, /* mGridWaitForPrimaryB */ 1
+, /* mHoistLoadTaskInit */ 1
+, /* mHoistMmaTaskTryWaits */ 0
+, /* mK */ 2048
+, /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
+, /* mM */ 256
+, /* mMmaK */ 64
+, /* mMmaKind */ trtllm::gen::MmaKind(4)
+, /* mMmaM */ 128
+, /* mMmaN */ 8
+, /* mMockAllReduce */ 0
+, /* mN */ 256
+, /* mNumSlicesForSplitK */ 1
+, /* mNumSlicesForSliceK */ 1
+, /* mNumStages */ 4
+, /* mNumStagesMma */ 1
+, /* mNumStagesMmaWithinWorkTile */ 1
+, /* mNumStagesMmaAcrossWorkTile */ 1
+, /* mNumStagesWorkId */ 3
+, /* mOutputDebugTensors */ 0
+, /* mUseShuffledMatrixA */ 1
+, /* mSliceK */ 0
+, /* mSplitK */ gemm::SplitK(0)
+, /* mTransposeMmaOutput */ 1
+, /* mTileM */ 128
+, /* mTileN */ 8
+, /* mTileK */ 512
+, /* mUseUnrollLoop2xForMma */ 1
+, /* mUseCustomMmaSchedule */ 1
+, /* mUseHoistTryWaitForCustomMmaSchedule */ 0
+, /* mUseDeepSeekFp8 */ 0
+, /* mUsePerTokenSfA */ 0
+, /* mUsePerTokenSfB */ 0
+, /* mUseTmaStore */ 1
+, /* mUseTwoTmaLoadWarps */ 1
+, /* mUseTwoMmaWarps */ 0
+, /* mSfLayoutA */ trtllm::gen::SfLayout(3)
+, /* mSfLayoutB */ trtllm::gen::SfLayout(0)
+, /* mSfLayoutC */ trtllm::gen::SfLayout(1)
+, /* mSfReshapeFactor */ 1
+, /* mTileScheduler */ gemm::TileScheduler(0)
+, /* mActType */ gemmGatedAct::ActType(0)
+, /* mBatchedM */ {}
+, /* mBatchedN */ {}
+, /* mBatchMode */ batchedGemm::BatchedGemmOptions::BatchMode(1)
+, /* mNumBatches */ 128
+, /* mIsStaticBatch */ 0
+, /* mNumTokens */ 32
+, /* mRouteImpl */ batchedGemm::RouteImpl(1)
+, /* mGridWaitForPrimaryRouting */ 1
+, /* mFusedAct */ 1
+, /* mNumRegsPerThreadNonEpilogueWarp */ 0
+, /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
+ }, gemm::SmVersion::Sm100a },
+{BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin, BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin_len, 165888, "BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a", 480, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
+, /* mClusterDimX */ 1
+, /* mClusterDimY */ 1
+, /* mClusterDimZ */ 1
+, /* mDtypeAcc */ trtllm::gen::Dtype(1056776)
+, /* mDtypeA */ trtllm::gen::Dtype(17826818)
+, /* mDtypeB */ trtllm::gen::Dtype(17826818)
+, /* mDtypeC */ trtllm::gen::Dtype(17826818)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(17826818)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(17826818)
+, /* mEnablesEarlyExit */ 1
+, /* mEnablesDelayedEarlyExit */ 0
+, /* mEnablesGlobalPtxKnobs */ 1
+, /* mEpilogueLdtmDps */ 16
+, /* mEpilogueLdtmBits */ 256
+, /* mEpilogueTileM */ 128
+, /* mEpilogueTileN */ 8
+, /* mGridTriggerSecondaryA */ 0
+, /* mGridTriggerSecondaryB */ 1
+, /* mGridWaitForPrimaryEarlyExit */ 1
+, /* mGridWaitForPrimaryA */ 0
+, /* mGridWaitForPrimaryB */ 1
+, /* mHoistLoadTaskInit */ 1
+, /* mHoistMmaTaskTryWaits */ 0
+, /* mK */ 2048
+, /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
+, /* mM */ 256
+, /* mMmaK */ 64
+, /* mMmaKind */ trtllm::gen::MmaKind(4)
+, /* mMmaM */ 128
+, /* mMmaN */ 8
+, /* mMockAllReduce */ 0
+, /* mN */ 256
+, /* mNumSlicesForSplitK */ 1
+, /* mNumSlicesForSliceK */ 1
+, /* mNumStages */ 4
+, /* mNumStagesMma */ 1
+, /* mNumStagesMmaWithinWorkTile */ 1
+, /* mNumStagesMmaAcrossWorkTile */ 1
+, /* mNumStagesWorkId */ 3
+, /* mOutputDebugTensors */ 0
+, /* mUseShuffledMatrixA */ 1
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
-, /* mTileN */ 32
-, /* mTileK */ 128
-, /* mUseUnrollLoop2xForMma */ 0
+, /* mTileN */ 8
+, /* mTileK */ 512
+, /* mUseUnrollLoop2xForMma */ 1
 , /* mUseCustomMmaSchedule */ 1
 , /* mUseHoistTryWaitForCustomMmaSchedule */ 0
-, /* mUseDeepSeekFp8 */ 1
+, /* mUseDeepSeekFp8 */ 0
 , /* mUsePerTokenSfA */ 0
 , /* mUsePerTokenSfB */ 0
 , /* mUseTmaStore */ 1
 , /* mUseTwoTmaLoadWarps */ 1
 , /* mUseTwoMmaWarps */ 0
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
-, /* mSfLayoutB */ trtllm::gen::SfLayout(3)
-, /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfLayoutB */ trtllm::gen::SfLayout(0)
+, /* mSfLayoutC */ trtllm::gen::SfLayout(1)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
 , /* mBatchedN */ {}
 , /* mBatchMode */ batchedGemm::BatchedGemmOptions::BatchMode(1)
-, /* mNumBatches */ 2
+, /* mNumBatches */ 128
 , /* mIsStaticBatch */ 0
-, /* mNumTokens */ 2
-, /* mRouteAct */ 1
+, /* mNumTokens */ 32
+, /* mRouteImpl */ batchedGemm::RouteImpl(1)
 , /* mGridWaitForPrimaryRouting */ 1
-, /* mFusedAct */ 0
+, /* mFusedAct */ 1
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x32x128u2_EpiTile64x32_Mma64x32x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x32x128u2_EpiTile64x32_Mma64x32x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin_len, 93184, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x32x128u2_EpiTile64x32_Mma64x32x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin, BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin_len, 165888, "BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a", 448, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
 , /* mDtypeAcc */ trtllm::gen::Dtype(1056776)
-, /* mDtypeA */ trtllm::gen::Dtype(1050629)
-, /* mDtypeB */ trtllm::gen::Dtype(1050629)
-, /* mDtypeC */ trtllm::gen::Dtype(1050629)
-, /* mEnablesEarlyExit */ 1
+, /* mDtypeA */ trtllm::gen::Dtype(17826818)
+, /* mDtypeB */ trtllm::gen::Dtype(17826818)
+, /* mDtypeC */ trtllm::gen::Dtype(17826818)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(17826818)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(17826818)
+, /* mEnablesEarlyExit */ 0
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
 , /* mEpilogueLdtmDps */ 16
 , /* mEpilogueLdtmBits */ 256
-, /* mEpilogueTileM */ 64
-, /* mEpilogueTileN */ 32
+, /* mEpilogueTileM */ 128
+, /* mEpilogueTileN */ 8
 , /* mGridTriggerSecondaryA */ 0
 , /* mGridTriggerSecondaryB */ 1
 , /* mGridWaitForPrimaryEarlyExit */ 1
 , /* mGridWaitForPrimaryA */ 0
 , /* mGridWaitForPrimaryB */ 1
 , /* mHoistLoadTaskInit */ 1
-, /* mHoistMmaTaskTryWaits */ 1
+, /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
-, /* mMmaK */ 32
-, /* mMmaKind */ trtllm::gen::MmaKind(2)
-, /* mMmaM */ 64
-, /* mMmaN */ 32
+, /* mMmaK */ 64
+, /* mMmaKind */ trtllm::gen::MmaKind(4)
+, /* mMmaM */ 128
+, /* mMmaN */ 8
 , /* mMockAllReduce */ 0
 , /* mN */ 256
 , /* mNumSlicesForSplitK */ 1
 , /* mNumSlicesForSliceK */ 1
 , /* mNumStages */ 4
-, /* mNumStagesMma */ 4
-, /* mNumStagesMmaWithinWorkTile */ 2
-, /* mNumStagesMmaAcrossWorkTile */ 2
+, /* mNumStagesMma */ 1
+, /* mNumStagesMmaWithinWorkTile */ 1
+, /* mNumStagesMmaAcrossWorkTile */ 1
 , /* mNumStagesWorkId */ 3
 , /* mOutputDebugTensors */ 0
-, /* mUseShuffledMatrixA */ 0
+, /* mUseShuffledMatrixA */ 1
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
-, /* mTileN */ 32
-, /* mTileK */ 128
+, /* mTileN */ 8
+, /* mTileK */ 512
 , /* mUseUnrollLoop2xForMma */ 1
 , /* mUseCustomMmaSchedule */ 1
 , /* mUseHoistTryWaitForCustomMmaSchedule */ 0
-, /* mUseDeepSeekFp8 */ 1
+, /* mUseDeepSeekFp8 */ 0
 , /* mUsePerTokenSfA */ 0
 , /* mUsePerTokenSfB */ 0
 , /* mUseTmaStore */ 1
 , /* mUseTwoTmaLoadWarps */ 1
 , /* mUseTwoMmaWarps */ 0
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
-, /* mSfLayoutB */ trtllm::gen::SfLayout(3)
-, /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfLayoutB */ trtllm::gen::SfLayout(1)
+, /* mSfLayoutC */ trtllm::gen::SfLayout(1)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
 , /* mBatchedN */ {}
 , /* mBatchMode */ batchedGemm::BatchedGemmOptions::BatchMode(1)
 , /* mNumBatches */ 2
-, /* mIsStaticBatch */ 0
-, /* mNumTokens */ 2
-, /* mRouteAct */ 1
+, /* mIsStaticBatch */ 1
+, /* mNumTokens */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 1
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x64x128_EpiTile64x64_Mma64x64x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x64x128_EpiTile64x64_Mma64x64x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin_len, 115712, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x64x128_EpiTile64x64_Mma64x64x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x128x128_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin, BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x128x128_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin_len, 158720, "BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x128x128_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -3898,13 +4856,15 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
 , /* mDtypeC */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
 , /* mEpilogueLdtmDps */ 16
 , /* mEpilogueLdtmBits */ 256
 , /* mEpilogueTileM */ 64
-, /* mEpilogueTileN */ 64
+, /* mEpilogueTileN */ 128
 , /* mGridTriggerSecondaryA */ 0
 , /* mGridTriggerSecondaryB */ 1
 , /* mGridWaitForPrimaryEarlyExit */ 1
@@ -3914,11 +4874,13 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 1
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
 , /* mMmaM */ 64
-, /* mMmaN */ 64
+, /* mMmaN */ 128
 , /* mMockAllReduce */ 0
 , /* mN */ 256
 , /* mNumSlicesForSplitK */ 1
@@ -3932,11 +4894,9 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseShuffledMatrixA */ 0
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
-, /* mTileN */ 64
+, /* mTileN */ 128
 , /* mTileK */ 128
 , /* mUseUnrollLoop2xForMma */ 0
 , /* mUseCustomMmaSchedule */ 1
@@ -3950,6 +4910,7 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -3958,13 +4919,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 1
+, /* mRouteImpl */ batchedGemm::RouteImpl(1)
 , /* mGridWaitForPrimaryRouting */ 1
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x64x128u2_EpiTile64x64_Mma64x64x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x64x128u2_EpiTile64x64_Mma64x64x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin_len, 115712, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x64x128u2_EpiTile64x64_Mma64x64x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x128x128u2_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin, BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x128x128u2_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin_len, 158720, "BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x128x128u2_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -3972,13 +4936,15 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
 , /* mDtypeC */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
 , /* mEpilogueLdtmDps */ 16
 , /* mEpilogueLdtmBits */ 256
 , /* mEpilogueTileM */ 64
-, /* mEpilogueTileN */ 64
+, /* mEpilogueTileN */ 128
 , /* mGridTriggerSecondaryA */ 0
 , /* mGridTriggerSecondaryB */ 1
 , /* mGridWaitForPrimaryEarlyExit */ 1
@@ -3988,11 +4954,13 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 1
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
 , /* mMmaM */ 64
-, /* mMmaN */ 64
+, /* mMmaN */ 128
 , /* mMockAllReduce */ 0
 , /* mN */ 256
 , /* mNumSlicesForSplitK */ 1
@@ -4006,11 +4974,9 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseShuffledMatrixA */ 0
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
-, /* mTileN */ 64
+, /* mTileN */ 128
 , /* mTileK */ 128
 , /* mUseUnrollLoop2xForMma */ 1
 , /* mUseCustomMmaSchedule */ 1
@@ -4024,6 +4990,7 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -4032,13 +4999,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 1
+, /* mRouteImpl */ batchedGemm::RouteImpl(1)
 , /* mGridWaitForPrimaryRouting */ 1
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x128_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x128_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin_len, 149504, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x128_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x16x128_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin, BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x16x128_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin_len, 82944, "BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x16x128_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -4046,45 +5016,47 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
 , /* mDtypeC */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
 , /* mEpilogueLdtmDps */ 16
 , /* mEpilogueLdtmBits */ 256
 , /* mEpilogueTileM */ 64
-, /* mEpilogueTileN */ 8
+, /* mEpilogueTileN */ 16
 , /* mGridTriggerSecondaryA */ 0
 , /* mGridTriggerSecondaryB */ 1
 , /* mGridWaitForPrimaryEarlyExit */ 1
 , /* mGridWaitForPrimaryA */ 0
 , /* mGridWaitForPrimaryB */ 1
 , /* mHoistLoadTaskInit */ 1
-, /* mHoistMmaTaskTryWaits */ 0
+, /* mHoistMmaTaskTryWaits */ 1
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
 , /* mMmaM */ 64
-, /* mMmaN */ 8
+, /* mMmaN */ 16
 , /* mMockAllReduce */ 0
 , /* mN */ 256
 , /* mNumSlicesForSplitK */ 1
 , /* mNumSlicesForSliceK */ 1
-, /* mNumStages */ 8
-, /* mNumStagesMma */ 2
+, /* mNumStages */ 4
+, /* mNumStagesMma */ 4
 , /* mNumStagesMmaWithinWorkTile */ 2
-, /* mNumStagesMmaAcrossWorkTile */ 1
+, /* mNumStagesMmaAcrossWorkTile */ 2
 , /* mNumStagesWorkId */ 3
 , /* mOutputDebugTensors */ 0
 , /* mUseShuffledMatrixA */ 0
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
-, /* mTileN */ 8
+, /* mTileN */ 16
 , /* mTileK */ 128
 , /* mUseUnrollLoop2xForMma */ 0
 , /* mUseCustomMmaSchedule */ 1
@@ -4098,6 +5070,7 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -4106,13 +5079,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 1
+, /* mRouteImpl */ batchedGemm::RouteImpl(1)
 , /* mGridWaitForPrimaryRouting */ 1
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x128_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x128_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_sm100a_cubin_len, 61440, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x128_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_sm100a", 384, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x16x128u2_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin, BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x16x128u2_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin_len, 82944, "BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x16x128u2_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -4120,47 +5096,49 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
 , /* mDtypeC */ trtllm::gen::Dtype(1050629)
-, /* mEnablesEarlyExit */ 0
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
+, /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
 , /* mEpilogueLdtmDps */ 16
 , /* mEpilogueLdtmBits */ 256
 , /* mEpilogueTileM */ 64
-, /* mEpilogueTileN */ 8
+, /* mEpilogueTileN */ 16
 , /* mGridTriggerSecondaryA */ 0
 , /* mGridTriggerSecondaryB */ 1
 , /* mGridWaitForPrimaryEarlyExit */ 1
 , /* mGridWaitForPrimaryA */ 0
 , /* mGridWaitForPrimaryB */ 1
 , /* mHoistLoadTaskInit */ 1
-, /* mHoistMmaTaskTryWaits */ 0
+, /* mHoistMmaTaskTryWaits */ 1
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
 , /* mMmaM */ 64
-, /* mMmaN */ 8
+, /* mMmaN */ 16
 , /* mMockAllReduce */ 0
 , /* mN */ 256
 , /* mNumSlicesForSplitK */ 1
 , /* mNumSlicesForSliceK */ 1
-, /* mNumStages */ 3
-, /* mNumStagesMma */ 2
+, /* mNumStages */ 4
+, /* mNumStagesMma */ 4
 , /* mNumStagesMmaWithinWorkTile */ 2
-, /* mNumStagesMmaAcrossWorkTile */ 1
+, /* mNumStagesMmaAcrossWorkTile */ 2
 , /* mNumStagesWorkId */ 3
 , /* mOutputDebugTensors */ 0
 , /* mUseShuffledMatrixA */ 0
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
-, /* mTileN */ 8
+, /* mTileN */ 16
 , /* mTileK */ 128
-, /* mUseUnrollLoop2xForMma */ 0
+, /* mUseUnrollLoop2xForMma */ 1
 , /* mUseCustomMmaSchedule */ 1
 , /* mUseHoistTryWaitForCustomMmaSchedule */ 0
 , /* mUseDeepSeekFp8 */ 1
@@ -4172,21 +5150,25 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
 , /* mBatchedN */ {}
 , /* mBatchMode */ batchedGemm::BatchedGemmOptions::BatchMode(1)
 , /* mNumBatches */ 2
-, /* mIsStaticBatch */ 1
-, /* mNumTokens */ 0
-, /* mRouteAct */ 0
+, /* mIsStaticBatch */ 0
+, /* mNumTokens */ 2
+, /* mRouteImpl */ batchedGemm::RouteImpl(1)
 , /* mGridWaitForPrimaryRouting */ 1
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x128u2_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x128u2_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a_cubin_len, 149504, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x128u2_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_InplaceRoute_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x32x128_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin, BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x32x128_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin_len, 93184, "BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x32x128_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -4194,47 +5176,49 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
 , /* mDtypeC */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
 , /* mEpilogueLdtmDps */ 16
 , /* mEpilogueLdtmBits */ 256
 , /* mEpilogueTileM */ 64
-, /* mEpilogueTileN */ 8
+, /* mEpilogueTileN */ 32
 , /* mGridTriggerSecondaryA */ 0
 , /* mGridTriggerSecondaryB */ 1
 , /* mGridWaitForPrimaryEarlyExit */ 1
 , /* mGridWaitForPrimaryA */ 0
 , /* mGridWaitForPrimaryB */ 1
 , /* mHoistLoadTaskInit */ 1
-, /* mHoistMmaTaskTryWaits */ 0
+, /* mHoistMmaTaskTryWaits */ 1
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
 , /* mMmaM */ 64
-, /* mMmaN */ 8
+, /* mMmaN */ 32
 , /* mMockAllReduce */ 0
 , /* mN */ 256
 , /* mNumSlicesForSplitK */ 1
 , /* mNumSlicesForSliceK */ 1
-, /* mNumStages */ 8
-, /* mNumStagesMma */ 2
+, /* mNumStages */ 4
+, /* mNumStagesMma */ 4
 , /* mNumStagesMmaWithinWorkTile */ 2
-, /* mNumStagesMmaAcrossWorkTile */ 1
+, /* mNumStagesMmaAcrossWorkTile */ 2
 , /* mNumStagesWorkId */ 3
 , /* mOutputDebugTensors */ 0
 , /* mUseShuffledMatrixA */ 0
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
-, /* mTileN */ 8
+, /* mTileN */ 32
 , /* mTileK */ 128
-, /* mUseUnrollLoop2xForMma */ 1
+, /* mUseUnrollLoop2xForMma */ 0
 , /* mUseCustomMmaSchedule */ 1
 , /* mUseHoistTryWaitForCustomMmaSchedule */ 0
 , /* mUseDeepSeekFp8 */ 1
@@ -4246,6 +5230,7 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -4254,13 +5239,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 1
+, /* mRouteImpl */ batchedGemm::RouteImpl(1)
 , /* mGridWaitForPrimaryRouting */ 1
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x128u2_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x128u2_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_sm100a_cubin_len, 61440, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x128u2_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_sm100a", 384, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x32x128u2_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin, BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x32x128u2_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin_len, 93184, "BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x32x128u2_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -4268,45 +5256,47 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
 , /* mDtypeC */ trtllm::gen::Dtype(1050629)
-, /* mEnablesEarlyExit */ 0
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
+, /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
 , /* mEpilogueLdtmDps */ 16
 , /* mEpilogueLdtmBits */ 256
 , /* mEpilogueTileM */ 64
-, /* mEpilogueTileN */ 8
+, /* mEpilogueTileN */ 32
 , /* mGridTriggerSecondaryA */ 0
 , /* mGridTriggerSecondaryB */ 1
 , /* mGridWaitForPrimaryEarlyExit */ 1
 , /* mGridWaitForPrimaryA */ 0
 , /* mGridWaitForPrimaryB */ 1
 , /* mHoistLoadTaskInit */ 1
-, /* mHoistMmaTaskTryWaits */ 0
+, /* mHoistMmaTaskTryWaits */ 1
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
 , /* mMmaM */ 64
-, /* mMmaN */ 8
+, /* mMmaN */ 32
 , /* mMockAllReduce */ 0
 , /* mN */ 256
 , /* mNumSlicesForSplitK */ 1
 , /* mNumSlicesForSliceK */ 1
-, /* mNumStages */ 3
-, /* mNumStagesMma */ 2
+, /* mNumStages */ 4
+, /* mNumStagesMma */ 4
 , /* mNumStagesMmaWithinWorkTile */ 2
-, /* mNumStagesMmaAcrossWorkTile */ 1
+, /* mNumStagesMmaAcrossWorkTile */ 2
 , /* mNumStagesWorkId */ 3
 , /* mOutputDebugTensors */ 0
 , /* mUseShuffledMatrixA */ 0
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
-, /* mTileN */ 8
+, /* mTileN */ 32
 , /* mTileK */ 128
 , /* mUseUnrollLoop2xForMma */ 1
 , /* mUseCustomMmaSchedule */ 1
@@ -4320,21 +5310,25 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
 , /* mBatchedN */ {}
 , /* mBatchMode */ batchedGemm::BatchedGemmOptions::BatchMode(1)
 , /* mNumBatches */ 2
-, /* mIsStaticBatch */ 1
-, /* mNumTokens */ 0
-, /* mRouteAct */ 0
+, /* mIsStaticBatch */ 0
+, /* mNumTokens */ 2
+, /* mRouteImpl */ batchedGemm::RouteImpl(1)
 , /* mGridWaitForPrimaryRouting */ 1
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_InplaceRoute_GatedAct_perTokenSfB_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_InplaceRoute_GatedAct_perTokenSfB_dynamic_sm100a_cubin_len, 216064, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_InplaceRoute_GatedAct_perTokenSfB_dynamic_sm100a", 256, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x64x128_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin, BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x64x128_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin_len, 115712, "BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x64x128_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -4342,58 +5336,61 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
 , /* mDtypeC */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
 , /* mEpilogueLdtmDps */ 16
 , /* mEpilogueLdtmBits */ 256
-, /* mEpilogueTileM */ 128
-, /* mEpilogueTileN */ 8
+, /* mEpilogueTileM */ 64
+, /* mEpilogueTileN */ 64
 , /* mGridTriggerSecondaryA */ 0
 , /* mGridTriggerSecondaryB */ 1
 , /* mGridWaitForPrimaryEarlyExit */ 1
 , /* mGridWaitForPrimaryA */ 0
 , /* mGridWaitForPrimaryB */ 1
 , /* mHoistLoadTaskInit */ 1
-, /* mHoistMmaTaskTryWaits */ 0
+, /* mHoistMmaTaskTryWaits */ 1
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
-, /* mMmaM */ 128
-, /* mMmaN */ 8
+, /* mMmaM */ 64
+, /* mMmaN */ 64
 , /* mMockAllReduce */ 0
 , /* mN */ 256
 , /* mNumSlicesForSplitK */ 1
 , /* mNumSlicesForSliceK */ 1
-, /* mNumStages */ 3
-, /* mNumStagesMma */ 1
-, /* mNumStagesMmaWithinWorkTile */ 1
-, /* mNumStagesMmaAcrossWorkTile */ 1
+, /* mNumStages */ 4
+, /* mNumStagesMma */ 4
+, /* mNumStagesMmaWithinWorkTile */ 2
+, /* mNumStagesMmaAcrossWorkTile */ 2
 , /* mNumStagesWorkId */ 3
 , /* mOutputDebugTensors */ 0
-, /* mUseShuffledMatrixA */ 1
+, /* mUseShuffledMatrixA */ 0
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
-, /* mTileN */ 8
-, /* mTileK */ 512
+, /* mTileN */ 64
+, /* mTileK */ 128
 , /* mUseUnrollLoop2xForMma */ 0
 , /* mUseCustomMmaSchedule */ 1
 , /* mUseHoistTryWaitForCustomMmaSchedule */ 0
-, /* mUseDeepSeekFp8 */ 0
+, /* mUseDeepSeekFp8 */ 1
 , /* mUsePerTokenSfA */ 0
-, /* mUsePerTokenSfB */ 1
+, /* mUsePerTokenSfB */ 0
 , /* mUseTmaStore */ 1
 , /* mUseTwoTmaLoadWarps */ 1
 , /* mUseTwoMmaWarps */ 0
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -4402,13 +5399,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 1
+, /* mRouteImpl */ batchedGemm::RouteImpl(1)
 , /* mGridWaitForPrimaryRouting */ 1
-, /* mFusedAct */ 1
+, /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_sm100a_cubin_len, 216064, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_sm100a", 224, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x64x128u2_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin, BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x64x128u2_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin_len, 115712, "BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x64x128u2_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -4416,50 +5416,52 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
 , /* mDtypeC */ trtllm::gen::Dtype(1050629)
-, /* mEnablesEarlyExit */ 0
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
+, /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
 , /* mEpilogueLdtmDps */ 16
 , /* mEpilogueLdtmBits */ 256
-, /* mEpilogueTileM */ 128
-, /* mEpilogueTileN */ 8
+, /* mEpilogueTileM */ 64
+, /* mEpilogueTileN */ 64
 , /* mGridTriggerSecondaryA */ 0
 , /* mGridTriggerSecondaryB */ 1
 , /* mGridWaitForPrimaryEarlyExit */ 1
 , /* mGridWaitForPrimaryA */ 0
 , /* mGridWaitForPrimaryB */ 1
 , /* mHoistLoadTaskInit */ 1
-, /* mHoistMmaTaskTryWaits */ 0
+, /* mHoistMmaTaskTryWaits */ 1
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
-, /* mMmaM */ 128
-, /* mMmaN */ 8
+, /* mMmaM */ 64
+, /* mMmaN */ 64
 , /* mMockAllReduce */ 0
 , /* mN */ 256
 , /* mNumSlicesForSplitK */ 1
 , /* mNumSlicesForSliceK */ 1
-, /* mNumStages */ 3
-, /* mNumStagesMma */ 1
-, /* mNumStagesMmaWithinWorkTile */ 1
-, /* mNumStagesMmaAcrossWorkTile */ 1
+, /* mNumStages */ 4
+, /* mNumStagesMma */ 4
+, /* mNumStagesMmaWithinWorkTile */ 2
+, /* mNumStagesMmaAcrossWorkTile */ 2
 , /* mNumStagesWorkId */ 3
 , /* mOutputDebugTensors */ 0
-, /* mUseShuffledMatrixA */ 1
+, /* mUseShuffledMatrixA */ 0
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
-, /* mTileN */ 8
-, /* mTileK */ 512
-, /* mUseUnrollLoop2xForMma */ 0
+, /* mTileN */ 64
+, /* mTileK */ 128
+, /* mUseUnrollLoop2xForMma */ 1
 , /* mUseCustomMmaSchedule */ 1
 , /* mUseHoistTryWaitForCustomMmaSchedule */ 0
-, /* mUseDeepSeekFp8 */ 0
+, /* mUseDeepSeekFp8 */ 1
 , /* mUsePerTokenSfA */ 0
 , /* mUsePerTokenSfB */ 0
 , /* mUseTmaStore */ 1
@@ -4468,21 +5470,25 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
 , /* mBatchedN */ {}
 , /* mBatchMode */ batchedGemm::BatchedGemmOptions::BatchMode(1)
 , /* mNumBatches */ 2
-, /* mIsStaticBatch */ 1
-, /* mNumTokens */ 0
-, /* mRouteAct */ 0
+, /* mIsStaticBatch */ 0
+, /* mNumTokens */ 2
+, /* mRouteImpl */ batchedGemm::RouteImpl(1)
 , /* mGridWaitForPrimaryRouting */ 1
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_InplaceRoute_GatedAct_perTokenSfB_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_InplaceRoute_GatedAct_perTokenSfB_dynamic_sm100a_cubin_len, 216064, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_InplaceRoute_GatedAct_perTokenSfB_dynamic_sm100a", 256, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin, BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin_len, 149504, "BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -4490,12 +5496,14 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
 , /* mDtypeC */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
 , /* mEpilogueLdtmDps */ 16
 , /* mEpilogueLdtmBits */ 256
-, /* mEpilogueTileM */ 128
+, /* mEpilogueTileM */ 64
 , /* mEpilogueTileN */ 8
 , /* mGridTriggerSecondaryA */ 0
 , /* mGridTriggerSecondaryB */ 1
@@ -4506,42 +5514,43 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
-, /* mMmaM */ 128
+, /* mMmaM */ 64
 , /* mMmaN */ 8
 , /* mMockAllReduce */ 0
 , /* mN */ 256
 , /* mNumSlicesForSplitK */ 1
 , /* mNumSlicesForSliceK */ 1
-, /* mNumStages */ 3
-, /* mNumStagesMma */ 1
-, /* mNumStagesMmaWithinWorkTile */ 1
+, /* mNumStages */ 8
+, /* mNumStagesMma */ 2
+, /* mNumStagesMmaWithinWorkTile */ 2
 , /* mNumStagesMmaAcrossWorkTile */ 1
 , /* mNumStagesWorkId */ 3
 , /* mOutputDebugTensors */ 0
-, /* mUseShuffledMatrixA */ 1
+, /* mUseShuffledMatrixA */ 0
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
-, /* mTileK */ 512
-, /* mUseUnrollLoop2xForMma */ 1
+, /* mTileK */ 128
+, /* mUseUnrollLoop2xForMma */ 0
 , /* mUseCustomMmaSchedule */ 1
 , /* mUseHoistTryWaitForCustomMmaSchedule */ 0
-, /* mUseDeepSeekFp8 */ 0
+, /* mUseDeepSeekFp8 */ 1
 , /* mUsePerTokenSfA */ 0
-, /* mUsePerTokenSfB */ 1
+, /* mUsePerTokenSfB */ 0
 , /* mUseTmaStore */ 1
 , /* mUseTwoTmaLoadWarps */ 1
 , /* mUseTwoMmaWarps */ 0
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -4550,13 +5559,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 1
+, /* mRouteImpl */ batchedGemm::RouteImpl(1)
 , /* mGridWaitForPrimaryRouting */ 1
-, /* mFusedAct */ 1
+, /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_sm100a_cubin_len, 216064, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_E4m3_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_sm100a", 224, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin, BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin_len, 61440, "BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a", 384, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -4564,12 +5576,14 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
 , /* mDtypeC */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
 , /* mEnablesEarlyExit */ 0
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
 , /* mEpilogueLdtmDps */ 16
 , /* mEpilogueLdtmBits */ 256
-, /* mEpilogueTileM */ 128
+, /* mEpilogueTileM */ 64
 , /* mEpilogueTileN */ 8
 , /* mGridTriggerSecondaryA */ 0
 , /* mGridTriggerSecondaryB */ 1
@@ -4580,34 +5594,34 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
-, /* mMmaM */ 128
+, /* mMmaM */ 64
 , /* mMmaN */ 8
 , /* mMockAllReduce */ 0
 , /* mN */ 256
 , /* mNumSlicesForSplitK */ 1
 , /* mNumSlicesForSliceK */ 1
 , /* mNumStages */ 3
-, /* mNumStagesMma */ 1
-, /* mNumStagesMmaWithinWorkTile */ 1
+, /* mNumStagesMma */ 2
+, /* mNumStagesMmaWithinWorkTile */ 2
 , /* mNumStagesMmaAcrossWorkTile */ 1
 , /* mNumStagesWorkId */ 3
 , /* mOutputDebugTensors */ 0
-, /* mUseShuffledMatrixA */ 1
+, /* mUseShuffledMatrixA */ 0
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
-, /* mTileK */ 512
-, /* mUseUnrollLoop2xForMma */ 1
+, /* mTileK */ 128
+, /* mUseUnrollLoop2xForMma */ 0
 , /* mUseCustomMmaSchedule */ 1
 , /* mUseHoistTryWaitForCustomMmaSchedule */ 0
-, /* mUseDeepSeekFp8 */ 0
+, /* mUseDeepSeekFp8 */ 1
 , /* mUsePerTokenSfA */ 0
 , /* mUsePerTokenSfB */ 0
 , /* mUseTmaStore */ 1
@@ -4616,6 +5630,7 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -4624,21 +5639,26 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 1
 , /* mNumTokens */ 0
-, /* mRouteAct */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 1
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Fp16_Tile128x8x128_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Fp16_Tile128x8x128_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_sm100a_cubin_len, 61440, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Fp16_Tile128x8x128_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_sm100a", 384, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin, BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin_len, 149504, "BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a", 416, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
 , /* mDtypeAcc */ trtllm::gen::Dtype(1056776)
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
-, /* mDtypeC */ trtllm::gen::Dtype(1052679)
-, /* mEnablesEarlyExit */ 0
+, /* mDtypeC */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
+, /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
 , /* mEpilogueLdtmDps */ 16
@@ -4654,6 +5674,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
@@ -4663,7 +5685,7 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mN */ 256
 , /* mNumSlicesForSplitK */ 1
 , /* mNumSlicesForSliceK */ 1
-, /* mNumStages */ 3
+, /* mNumStages */ 8
 , /* mNumStagesMma */ 2
 , /* mNumStagesMmaWithinWorkTile */ 2
 , /* mNumStagesMmaAcrossWorkTile */ 1
@@ -4672,13 +5694,11 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseShuffledMatrixA */ 0
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
 , /* mTileK */ 128
-, /* mUseUnrollLoop2xForMma */ 0
+, /* mUseUnrollLoop2xForMma */ 1
 , /* mUseCustomMmaSchedule */ 1
 , /* mUseHoistTryWaitForCustomMmaSchedule */ 0
 , /* mUseDeepSeekFp8 */ 1
@@ -4690,28 +5710,34 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
 , /* mBatchedN */ {}
 , /* mBatchMode */ batchedGemm::BatchedGemmOptions::BatchMode(1)
 , /* mNumBatches */ 2
-, /* mIsStaticBatch */ 1
-, /* mNumTokens */ 0
-, /* mRouteAct */ 0
+, /* mIsStaticBatch */ 0
+, /* mNumTokens */ 2
+, /* mRouteImpl */ batchedGemm::RouteImpl(1)
 , /* mGridWaitForPrimaryRouting */ 1
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Fp16_Tile128x8x128u2_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Fp16_Tile128x8x128u2_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_sm100a_cubin_len, 61440, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Fp16_Tile128x8x128u2_EpiTile64x8_Mma64x8x32_Cluster1x1x1_transposeMmaOutput_DsFp8_sm100a", 384, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin, BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin_len, 61440, "BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a", 384, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
 , /* mDtypeAcc */ trtllm::gen::Dtype(1056776)
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
-, /* mDtypeC */ trtllm::gen::Dtype(1052679)
+, /* mDtypeC */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
 , /* mEnablesEarlyExit */ 0
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -4728,6 +5754,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
@@ -4746,8 +5774,6 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseShuffledMatrixA */ 0
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
@@ -4764,6 +5790,7 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -4772,21 +5799,26 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 1
 , /* mNumTokens */ 0
-, /* mRouteAct */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 1
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Fp16_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Fp16_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_sm100a_cubin_len, 217088, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Fp16_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_sm100a", 224, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin, BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin_len, 216064, "BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a", 256, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
 , /* mDtypeAcc */ trtllm::gen::Dtype(1056776)
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
-, /* mDtypeC */ trtllm::gen::Dtype(1052679)
-, /* mEnablesEarlyExit */ 0
+, /* mDtypeC */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
+, /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
 , /* mEpilogueLdtmDps */ 16
@@ -4802,6 +5834,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
@@ -4820,8 +5854,6 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseShuffledMatrixA */ 1
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
@@ -4838,28 +5870,34 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
 , /* mBatchedN */ {}
 , /* mBatchMode */ batchedGemm::BatchedGemmOptions::BatchMode(1)
 , /* mNumBatches */ 2
-, /* mIsStaticBatch */ 1
-, /* mNumTokens */ 0
-, /* mRouteAct */ 0
+, /* mIsStaticBatch */ 0
+, /* mNumTokens */ 2
+, /* mRouteImpl */ batchedGemm::RouteImpl(1)
 , /* mGridWaitForPrimaryRouting */ 1
-, /* mFusedAct */ 0
+, /* mFusedAct */ 1
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Fp16_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_sm100a_cubin, BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Fp16_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_sm100a_cubin_len, 217088, "BatchedGemmKernel_BatchN_E4m3E4m3Fp32_Fp16_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_sm100a", 224, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin, BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin_len, 216064, "BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a", 224, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
 , /* mDtypeAcc */ trtllm::gen::Dtype(1056776)
 , /* mDtypeA */ trtllm::gen::Dtype(1050629)
 , /* mDtypeB */ trtllm::gen::Dtype(1050629)
-, /* mDtypeC */ trtllm::gen::Dtype(1052679)
+, /* mDtypeC */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
 , /* mEnablesEarlyExit */ 0
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -4876,6 +5914,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(2)
@@ -4894,13 +5934,11 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseShuffledMatrixA */ 1
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
 , /* mTileK */ 512
-, /* mUseUnrollLoop2xForMma */ 1
+, /* mUseUnrollLoop2xForMma */ 0
 , /* mUseCustomMmaSchedule */ 1
 , /* mUseHoistTryWaitForCustomMmaSchedule */ 0
 , /* mUseDeepSeekFp8 */ 0
@@ -4912,6 +5950,7 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -4920,20 +5959,25 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 1
 , /* mNumTokens */ 0
-, /* mRouteAct */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
 , /* mGridWaitForPrimaryRouting */ 1
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x256_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x256_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin_len, 153600, "BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x256_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a", 448, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin, BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin_len, 216064, "BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a", 256, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
 , /* mDtypeAcc */ trtllm::gen::Dtype(1056776)
-, /* mDtypeA */ trtllm::gen::Dtype(17826828)
-, /* mDtypeB */ trtllm::gen::Dtype(17827853)
-, /* mDtypeC */ trtllm::gen::Dtype(1052672)
+, /* mDtypeA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeB */ trtllm::gen::Dtype(1050629)
+, /* mDtypeC */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -4950,31 +5994,31 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
-, /* mMmaKind */ trtllm::gen::MmaKind(5)
+, /* mMmaKind */ trtllm::gen::MmaKind(2)
 , /* mMmaM */ 128
 , /* mMmaN */ 8
 , /* mMockAllReduce */ 0
 , /* mN */ 256
 , /* mNumSlicesForSplitK */ 1
 , /* mNumSlicesForSliceK */ 1
-, /* mNumStages */ 4
-, /* mNumStagesMma */ 2
+, /* mNumStages */ 3
+, /* mNumStagesMma */ 1
 , /* mNumStagesMmaWithinWorkTile */ 1
-, /* mNumStagesMmaAcrossWorkTile */ 2
+, /* mNumStagesMmaAcrossWorkTile */ 1
 , /* mNumStagesWorkId */ 3
 , /* mOutputDebugTensors */ 0
 , /* mUseShuffledMatrixA */ 1
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
-, /* mTileK */ 256
-, /* mUseUnrollLoop2xForMma */ 0
+, /* mTileK */ 512
+, /* mUseUnrollLoop2xForMma */ 1
 , /* mUseCustomMmaSchedule */ 1
 , /* mUseHoistTryWaitForCustomMmaSchedule */ 0
 , /* mUseDeepSeekFp8 */ 0
@@ -4984,8 +6028,9 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseTwoTmaLoadWarps */ 1
 , /* mUseTwoMmaWarps */ 0
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
-, /* mSfLayoutB */ trtllm::gen::SfLayout(1)
-, /* mSfLayoutC */ trtllm::gen::SfLayout(1)
+, /* mSfLayoutB */ trtllm::gen::SfLayout(3)
+, /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -4994,21 +6039,26 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 0
-, /* mGridWaitForPrimaryRouting */ 0
-, /* mFusedAct */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(1)
+, /* mGridWaitForPrimaryRouting */ 1
+, /* mFusedAct */ 1
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x256_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x256_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin_len, 153600, "BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x256_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a", 480, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin, BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin_len, 216064, "BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a", 224, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
 , /* mDtypeAcc */ trtllm::gen::Dtype(1056776)
-, /* mDtypeA */ trtllm::gen::Dtype(17826828)
-, /* mDtypeB */ trtllm::gen::Dtype(17827853)
-, /* mDtypeC */ trtllm::gen::Dtype(1052672)
-, /* mEnablesEarlyExit */ 1
+, /* mDtypeA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeB */ trtllm::gen::Dtype(1050629)
+, /* mDtypeC */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
+, /* mEnablesEarlyExit */ 0
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
 , /* mEpilogueLdtmDps */ 16
@@ -5024,31 +6074,31 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
-, /* mMmaKind */ trtllm::gen::MmaKind(5)
+, /* mMmaKind */ trtllm::gen::MmaKind(2)
 , /* mMmaM */ 128
 , /* mMmaN */ 8
 , /* mMockAllReduce */ 0
 , /* mN */ 256
 , /* mNumSlicesForSplitK */ 1
 , /* mNumSlicesForSliceK */ 1
-, /* mNumStages */ 4
-, /* mNumStagesMma */ 2
+, /* mNumStages */ 3
+, /* mNumStagesMma */ 1
 , /* mNumStagesMmaWithinWorkTile */ 1
-, /* mNumStagesMmaAcrossWorkTile */ 2
+, /* mNumStagesMmaAcrossWorkTile */ 1
 , /* mNumStagesWorkId */ 3
 , /* mOutputDebugTensors */ 0
 , /* mUseShuffledMatrixA */ 1
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
-, /* mTileK */ 256
-, /* mUseUnrollLoop2xForMma */ 0
+, /* mTileK */ 512
+, /* mUseUnrollLoop2xForMma */ 1
 , /* mUseCustomMmaSchedule */ 1
 , /* mUseHoistTryWaitForCustomMmaSchedule */ 0
 , /* mUseDeepSeekFp8 */ 0
@@ -5058,31 +6108,37 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseTwoTmaLoadWarps */ 1
 , /* mUseTwoMmaWarps */ 0
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
-, /* mSfLayoutB */ trtllm::gen::SfLayout(1)
-, /* mSfLayoutC */ trtllm::gen::SfLayout(1)
-, /* mTileScheduler */ gemm::TileScheduler(1)
+, /* mSfLayoutB */ trtllm::gen::SfLayout(3)
+, /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
+, /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
 , /* mBatchedN */ {}
 , /* mBatchMode */ batchedGemm::BatchedGemmOptions::BatchMode(1)
 , /* mNumBatches */ 2
-, /* mIsStaticBatch */ 0
-, /* mNumTokens */ 2
-, /* mRouteAct */ 0
-, /* mGridWaitForPrimaryRouting */ 0
+, /* mIsStaticBatch */ 1
+, /* mNumTokens */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
+, /* mGridWaitForPrimaryRouting */ 1
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x256u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x256u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin_len, 153600, "BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x256u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a", 448, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Fp16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin, BatchedGemmKernel_Fp16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin_len, 166912, "BatchedGemmKernel_Fp16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a", 448, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
 , /* mDtypeAcc */ trtllm::gen::Dtype(1056776)
-, /* mDtypeA */ trtllm::gen::Dtype(17826828)
-, /* mDtypeB */ trtllm::gen::Dtype(17827853)
-, /* mDtypeC */ trtllm::gen::Dtype(1052672)
-, /* mEnablesEarlyExit */ 1
+, /* mDtypeA */ trtllm::gen::Dtype(17826818)
+, /* mDtypeB */ trtllm::gen::Dtype(17826818)
+, /* mDtypeC */ trtllm::gen::Dtype(1052679)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(17826818)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(17826818)
+, /* mEnablesEarlyExit */ 0
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
 , /* mEpilogueLdtmDps */ 16
@@ -5098,9 +6154,11 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
-, /* mMmaK */ 32
-, /* mMmaKind */ trtllm::gen::MmaKind(5)
+, /* mMmaK */ 64
+, /* mMmaKind */ trtllm::gen::MmaKind(4)
 , /* mMmaM */ 128
 , /* mMmaN */ 8
 , /* mMockAllReduce */ 0
@@ -5108,21 +6166,19 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumSlicesForSplitK */ 1
 , /* mNumSlicesForSliceK */ 1
 , /* mNumStages */ 4
-, /* mNumStagesMma */ 2
+, /* mNumStagesMma */ 1
 , /* mNumStagesMmaWithinWorkTile */ 1
-, /* mNumStagesMmaAcrossWorkTile */ 2
+, /* mNumStagesMmaAcrossWorkTile */ 1
 , /* mNumStagesWorkId */ 3
 , /* mOutputDebugTensors */ 0
 , /* mUseShuffledMatrixA */ 1
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
-, /* mTileK */ 256
-, /* mUseUnrollLoop2xForMma */ 1
+, /* mTileK */ 512
+, /* mUseUnrollLoop2xForMma */ 0
 , /* mUseCustomMmaSchedule */ 1
 , /* mUseHoistTryWaitForCustomMmaSchedule */ 0
 , /* mUseDeepSeekFp8 */ 0
@@ -5134,29 +6190,35 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(1)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(1)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
 , /* mBatchedN */ {}
 , /* mBatchMode */ batchedGemm::BatchedGemmOptions::BatchMode(1)
 , /* mNumBatches */ 2
-, /* mIsStaticBatch */ 0
-, /* mNumTokens */ 2
-, /* mRouteAct */ 0
-, /* mGridWaitForPrimaryRouting */ 0
+, /* mIsStaticBatch */ 1
+, /* mNumTokens */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
+, /* mGridWaitForPrimaryRouting */ 1
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x256u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x256u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin_len, 153600, "BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x256u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a", 480, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Fp16_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin, BatchedGemmKernel_Fp16_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin_len, 166912, "BatchedGemmKernel_Fp16_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a", 448, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
 , /* mDtypeAcc */ trtllm::gen::Dtype(1056776)
-, /* mDtypeA */ trtllm::gen::Dtype(17826828)
-, /* mDtypeB */ trtllm::gen::Dtype(17827853)
-, /* mDtypeC */ trtllm::gen::Dtype(1052672)
-, /* mEnablesEarlyExit */ 1
+, /* mDtypeA */ trtllm::gen::Dtype(17826818)
+, /* mDtypeB */ trtllm::gen::Dtype(17826818)
+, /* mDtypeC */ trtllm::gen::Dtype(1052679)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(17826818)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(17826818)
+, /* mEnablesEarlyExit */ 0
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
 , /* mEpilogueLdtmDps */ 16
@@ -5172,9 +6234,11 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
-, /* mMmaK */ 32
-, /* mMmaKind */ trtllm::gen::MmaKind(5)
+, /* mMmaK */ 64
+, /* mMmaKind */ trtllm::gen::MmaKind(4)
 , /* mMmaM */ 128
 , /* mMmaN */ 8
 , /* mMockAllReduce */ 0
@@ -5182,20 +6246,18 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumSlicesForSplitK */ 1
 , /* mNumSlicesForSliceK */ 1
 , /* mNumStages */ 4
-, /* mNumStagesMma */ 2
+, /* mNumStagesMma */ 1
 , /* mNumStagesMmaWithinWorkTile */ 1
-, /* mNumStagesMmaAcrossWorkTile */ 2
+, /* mNumStagesMmaAcrossWorkTile */ 1
 , /* mNumStagesWorkId */ 3
 , /* mOutputDebugTensors */ 0
 , /* mUseShuffledMatrixA */ 1
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
-, /* mTileK */ 256
+, /* mTileK */ 512
 , /* mUseUnrollLoop2xForMma */ 1
 , /* mUseCustomMmaSchedule */ 1
 , /* mUseHoistTryWaitForCustomMmaSchedule */ 0
@@ -5208,34 +6270,40 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
 , /* mSfLayoutB */ trtllm::gen::SfLayout(1)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(1)
-, /* mTileScheduler */ gemm::TileScheduler(1)
+, /* mSfReshapeFactor */ 1
+, /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
 , /* mBatchedN */ {}
 , /* mBatchMode */ batchedGemm::BatchedGemmOptions::BatchMode(1)
 , /* mNumBatches */ 2
-, /* mIsStaticBatch */ 0
-, /* mNumTokens */ 2
-, /* mRouteAct */ 0
-, /* mGridWaitForPrimaryRouting */ 0
+, /* mIsStaticBatch */ 1
+, /* mNumTokens */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
+, /* mGridWaitForPrimaryRouting */ 1
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin_len, 225280, "BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a", 448, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Fp16_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin, BatchedGemmKernel_Fp16_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin_len, 61440, "BatchedGemmKernel_Fp16_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a", 384, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
 , /* mDtypeAcc */ trtllm::gen::Dtype(1056776)
-, /* mDtypeA */ trtllm::gen::Dtype(17826828)
-, /* mDtypeB */ trtllm::gen::Dtype(17827853)
-, /* mDtypeC */ trtllm::gen::Dtype(1052672)
-, /* mEnablesEarlyExit */ 1
+, /* mDtypeA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeB */ trtllm::gen::Dtype(1050629)
+, /* mDtypeC */ trtllm::gen::Dtype(1052679)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
+, /* mEnablesEarlyExit */ 0
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
 , /* mEpilogueLdtmDps */ 16
 , /* mEpilogueLdtmBits */ 256
-, /* mEpilogueTileM */ 128
+, /* mEpilogueTileM */ 64
 , /* mEpilogueTileN */ 8
 , /* mGridTriggerSecondaryA */ 0
 , /* mGridTriggerSecondaryB */ 1
@@ -5246,10 +6314,12 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
-, /* mMmaKind */ trtllm::gen::MmaKind(5)
-, /* mMmaM */ 128
+, /* mMmaKind */ trtllm::gen::MmaKind(2)
+, /* mMmaM */ 64
 , /* mMmaN */ 8
 , /* mMockAllReduce */ 0
 , /* mN */ 256
@@ -5257,59 +6327,63 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumSlicesForSliceK */ 1
 , /* mNumStages */ 3
 , /* mNumStagesMma */ 2
-, /* mNumStagesMmaWithinWorkTile */ 1
-, /* mNumStagesMmaAcrossWorkTile */ 2
+, /* mNumStagesMmaWithinWorkTile */ 2
+, /* mNumStagesMmaAcrossWorkTile */ 1
 , /* mNumStagesWorkId */ 3
 , /* mOutputDebugTensors */ 0
-, /* mUseShuffledMatrixA */ 1
+, /* mUseShuffledMatrixA */ 0
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
-, /* mTileK */ 512
+, /* mTileK */ 128
 , /* mUseUnrollLoop2xForMma */ 0
 , /* mUseCustomMmaSchedule */ 1
 , /* mUseHoistTryWaitForCustomMmaSchedule */ 0
-, /* mUseDeepSeekFp8 */ 0
+, /* mUseDeepSeekFp8 */ 1
 , /* mUsePerTokenSfA */ 0
 , /* mUsePerTokenSfB */ 0
 , /* mUseTmaStore */ 1
 , /* mUseTwoTmaLoadWarps */ 1
 , /* mUseTwoMmaWarps */ 0
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
-, /* mSfLayoutB */ trtllm::gen::SfLayout(1)
-, /* mSfLayoutC */ trtllm::gen::SfLayout(1)
+, /* mSfLayoutB */ trtllm::gen::SfLayout(3)
+, /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
 , /* mBatchedN */ {}
 , /* mBatchMode */ batchedGemm::BatchedGemmOptions::BatchMode(1)
 , /* mNumBatches */ 2
-, /* mIsStaticBatch */ 0
-, /* mNumTokens */ 2
-, /* mRouteAct */ 0
-, /* mGridWaitForPrimaryRouting */ 0
+, /* mIsStaticBatch */ 1
+, /* mNumTokens */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
+, /* mGridWaitForPrimaryRouting */ 1
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin_len, 225280, "BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a", 480, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Fp16_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin, BatchedGemmKernel_Fp16_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin_len, 61440, "BatchedGemmKernel_Fp16_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a", 384, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
 , /* mDtypeAcc */ trtllm::gen::Dtype(1056776)
-, /* mDtypeA */ trtllm::gen::Dtype(17826828)
-, /* mDtypeB */ trtllm::gen::Dtype(17827853)
-, /* mDtypeC */ trtllm::gen::Dtype(1052672)
-, /* mEnablesEarlyExit */ 1
+, /* mDtypeA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeB */ trtllm::gen::Dtype(1050629)
+, /* mDtypeC */ trtllm::gen::Dtype(1052679)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
+, /* mEnablesEarlyExit */ 0
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
 , /* mEpilogueLdtmDps */ 16
 , /* mEpilogueLdtmBits */ 256
-, /* mEpilogueTileM */ 128
+, /* mEpilogueTileM */ 64
 , /* mEpilogueTileN */ 8
 , /* mGridTriggerSecondaryA */ 0
 , /* mGridTriggerSecondaryB */ 1
@@ -5320,10 +6394,12 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
-, /* mMmaKind */ trtllm::gen::MmaKind(5)
-, /* mMmaM */ 128
+, /* mMmaKind */ trtllm::gen::MmaKind(2)
+, /* mMmaM */ 64
 , /* mMmaN */ 8
 , /* mMockAllReduce */ 0
 , /* mN */ 256
@@ -5331,54 +6407,58 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumSlicesForSliceK */ 1
 , /* mNumStages */ 3
 , /* mNumStagesMma */ 2
-, /* mNumStagesMmaWithinWorkTile */ 1
-, /* mNumStagesMmaAcrossWorkTile */ 2
+, /* mNumStagesMmaWithinWorkTile */ 2
+, /* mNumStagesMmaAcrossWorkTile */ 1
 , /* mNumStagesWorkId */ 3
 , /* mOutputDebugTensors */ 0
-, /* mUseShuffledMatrixA */ 1
+, /* mUseShuffledMatrixA */ 0
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
-, /* mTileK */ 512
-, /* mUseUnrollLoop2xForMma */ 0
+, /* mTileK */ 128
+, /* mUseUnrollLoop2xForMma */ 1
 , /* mUseCustomMmaSchedule */ 1
 , /* mUseHoistTryWaitForCustomMmaSchedule */ 0
-, /* mUseDeepSeekFp8 */ 0
+, /* mUseDeepSeekFp8 */ 1
 , /* mUsePerTokenSfA */ 0
 , /* mUsePerTokenSfB */ 0
 , /* mUseTmaStore */ 1
 , /* mUseTwoTmaLoadWarps */ 1
 , /* mUseTwoMmaWarps */ 0
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
-, /* mSfLayoutB */ trtllm::gen::SfLayout(1)
-, /* mSfLayoutC */ trtllm::gen::SfLayout(1)
-, /* mTileScheduler */ gemm::TileScheduler(1)
+, /* mSfLayoutB */ trtllm::gen::SfLayout(3)
+, /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
+, /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
 , /* mBatchedN */ {}
 , /* mBatchMode */ batchedGemm::BatchedGemmOptions::BatchMode(1)
 , /* mNumBatches */ 2
-, /* mIsStaticBatch */ 0
-, /* mNumTokens */ 2
-, /* mRouteAct */ 0
-, /* mGridWaitForPrimaryRouting */ 0
+, /* mIsStaticBatch */ 1
+, /* mNumTokens */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
+, /* mGridWaitForPrimaryRouting */ 1
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a_cubin_len, 225280, "BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_dynamic_sm100a", 448, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Fp16_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin, BatchedGemmKernel_Fp16_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin_len, 217088, "BatchedGemmKernel_Fp16_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a", 224, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
 , /* mDtypeAcc */ trtllm::gen::Dtype(1056776)
-, /* mDtypeA */ trtllm::gen::Dtype(17826828)
-, /* mDtypeB */ trtllm::gen::Dtype(17827853)
-, /* mDtypeC */ trtllm::gen::Dtype(1052672)
-, /* mEnablesEarlyExit */ 1
+, /* mDtypeA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeB */ trtllm::gen::Dtype(1050629)
+, /* mDtypeC */ trtllm::gen::Dtype(1052679)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
+, /* mEnablesEarlyExit */ 0
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
 , /* mEpilogueLdtmDps */ 16
@@ -5394,9 +6474,11 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
-, /* mMmaKind */ trtllm::gen::MmaKind(5)
+, /* mMmaKind */ trtllm::gen::MmaKind(2)
 , /* mMmaM */ 128
 , /* mMmaN */ 8
 , /* mMockAllReduce */ 0
@@ -5404,21 +6486,19 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumSlicesForSplitK */ 1
 , /* mNumSlicesForSliceK */ 1
 , /* mNumStages */ 3
-, /* mNumStagesMma */ 2
+, /* mNumStagesMma */ 1
 , /* mNumStagesMmaWithinWorkTile */ 1
-, /* mNumStagesMmaAcrossWorkTile */ 2
+, /* mNumStagesMmaAcrossWorkTile */ 1
 , /* mNumStagesWorkId */ 3
 , /* mOutputDebugTensors */ 0
 , /* mUseShuffledMatrixA */ 1
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
 , /* mTileK */ 512
-, /* mUseUnrollLoop2xForMma */ 1
+, /* mUseUnrollLoop2xForMma */ 0
 , /* mUseCustomMmaSchedule */ 1
 , /* mUseHoistTryWaitForCustomMmaSchedule */ 0
 , /* mUseDeepSeekFp8 */ 0
@@ -5428,31 +6508,37 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseTwoTmaLoadWarps */ 1
 , /* mUseTwoMmaWarps */ 0
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
-, /* mSfLayoutB */ trtllm::gen::SfLayout(1)
-, /* mSfLayoutC */ trtllm::gen::SfLayout(1)
+, /* mSfLayoutB */ trtllm::gen::SfLayout(3)
+, /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
 , /* mBatchedN */ {}
 , /* mBatchMode */ batchedGemm::BatchedGemmOptions::BatchMode(1)
 , /* mNumBatches */ 2
-, /* mIsStaticBatch */ 0
-, /* mNumTokens */ 2
-, /* mRouteAct */ 0
-, /* mGridWaitForPrimaryRouting */ 0
+, /* mIsStaticBatch */ 1
+, /* mNumTokens */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
+, /* mGridWaitForPrimaryRouting */ 1
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a_cubin_len, 225280, "BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_Bfloat16_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_persistent_dynamic_sm100a", 480, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_Fp16_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin, BatchedGemmKernel_Fp16_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin_len, 217088, "BatchedGemmKernel_Fp16_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a", 224, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
 , /* mDtypeAcc */ trtllm::gen::Dtype(1056776)
-, /* mDtypeA */ trtllm::gen::Dtype(17826828)
-, /* mDtypeB */ trtllm::gen::Dtype(17827853)
-, /* mDtypeC */ trtllm::gen::Dtype(1052672)
-, /* mEnablesEarlyExit */ 1
+, /* mDtypeA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeB */ trtllm::gen::Dtype(1050629)
+, /* mDtypeC */ trtllm::gen::Dtype(1052679)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(1050629)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(1050629)
+, /* mEnablesEarlyExit */ 0
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
 , /* mEpilogueLdtmDps */ 16
@@ -5468,9 +6554,11 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
-, /* mMmaKind */ trtllm::gen::MmaKind(5)
+, /* mMmaKind */ trtllm::gen::MmaKind(2)
 , /* mMmaM */ 128
 , /* mMmaN */ 8
 , /* mMockAllReduce */ 0
@@ -5478,16 +6566,14 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumSlicesForSplitK */ 1
 , /* mNumSlicesForSliceK */ 1
 , /* mNumStages */ 3
-, /* mNumStagesMma */ 2
+, /* mNumStagesMma */ 1
 , /* mNumStagesMmaWithinWorkTile */ 1
-, /* mNumStagesMmaAcrossWorkTile */ 2
+, /* mNumStagesMmaAcrossWorkTile */ 1
 , /* mNumStagesWorkId */ 3
 , /* mOutputDebugTensors */ 0
 , /* mUseShuffledMatrixA */ 1
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
@@ -5502,23 +6588,27 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseTwoTmaLoadWarps */ 1
 , /* mUseTwoMmaWarps */ 0
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
-, /* mSfLayoutB */ trtllm::gen::SfLayout(1)
-, /* mSfLayoutC */ trtllm::gen::SfLayout(1)
-, /* mTileScheduler */ gemm::TileScheduler(1)
+, /* mSfLayoutB */ trtllm::gen::SfLayout(3)
+, /* mSfLayoutC */ trtllm::gen::SfLayout(3)
+, /* mSfReshapeFactor */ 1
+, /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
 , /* mBatchedN */ {}
 , /* mBatchMode */ batchedGemm::BatchedGemmOptions::BatchMode(1)
 , /* mNumBatches */ 2
-, /* mIsStaticBatch */ 0
-, /* mNumTokens */ 2
-, /* mRouteAct */ 0
-, /* mGridWaitForPrimaryRouting */ 0
+, /* mIsStaticBatch */ 1
+, /* mNumTokens */ 0
+, /* mRouteImpl */ batchedGemm::RouteImpl(0)
+, /* mGridWaitForPrimaryRouting */ 1
 , /* mFusedAct */ 0
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_MxE4m3_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_InplaceRoute_GatedAct_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_MxE4m3_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_InplaceRoute_GatedAct_dynamic_sm100a_cubin_len, 224256, "BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_MxE4m3_Tile128x8x512_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_InplaceRoute_GatedAct_dynamic_sm100a", 480, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_MxE4m3_MxE2m1MxE4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin, BatchedGemmKernel_MxE4m3_MxE2m1MxE4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin_len, 224256, "BatchedGemmKernel_MxE4m3_MxE2m1MxE4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a", 480, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -5526,6 +6616,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(17826828)
 , /* mDtypeB */ trtllm::gen::Dtype(17827853)
 , /* mDtypeC */ trtllm::gen::Dtype(17827853)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(17826828)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(17827853)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -5542,6 +6634,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(5)
@@ -5560,8 +6654,6 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseShuffledMatrixA */ 1
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
@@ -5576,8 +6668,9 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseTwoTmaLoadWarps */ 1
 , /* mUseTwoMmaWarps */ 0
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
-, /* mSfLayoutB */ trtllm::gen::SfLayout(1)
+, /* mSfLayoutB */ trtllm::gen::SfLayout(0)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(1)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -5586,13 +6679,16 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 1
+, /* mRouteImpl */ batchedGemm::RouteImpl(1)
 , /* mGridWaitForPrimaryRouting */ 1
 , /* mFusedAct */ 1
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
-{BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_MxE4m3_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_InplaceRoute_GatedAct_dynamic_sm100a_cubin, BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_MxE4m3_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_InplaceRoute_GatedAct_dynamic_sm100a_cubin_len, 224256, "BatchedGemmKernel_BatchN_MxE2m1MxE4m3Fp32_MxE4m3_Tile128x8x512u2_EpiTile128x8_Mma128x8x32_Cluster1x1x1_transposeMmaOutput_InplaceRoute_GatedAct_dynamic_sm100a", 480, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+{BatchedGemmKernel_MxE4m3_MxE2m1MxE4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin, BatchedGemmKernel_MxE4m3_MxE2m1MxE4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin_len, 224256, "BatchedGemmKernel_MxE4m3_MxE2m1MxE4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a", 480, { /* mAllReduceAlgo */ gemm::AllReduceAlgo(0)
+, /* mBiasType */ gemm::BiasType(0)
+, /* mBlockK */ -1
 , /* mClusterDimX */ 1
 , /* mClusterDimY */ 1
 , /* mClusterDimZ */ 1
@@ -5600,6 +6696,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mDtypeA */ trtllm::gen::Dtype(17826828)
 , /* mDtypeB */ trtllm::gen::Dtype(17827853)
 , /* mDtypeC */ trtllm::gen::Dtype(17827853)
+, /* mDtypeMmaA */ trtllm::gen::Dtype(17826828)
+, /* mDtypeMmaB */ trtllm::gen::Dtype(17827853)
 , /* mEnablesEarlyExit */ 1
 , /* mEnablesDelayedEarlyExit */ 0
 , /* mEnablesGlobalPtxKnobs */ 1
@@ -5616,6 +6714,8 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mHoistMmaTaskTryWaits */ 0
 , /* mK */ 2048
 , /* mKernelTraits */ {}
+, /* mLayoutA */ gemm::MatrixLayout(0)
+, /* mLayoutB */ gemm::MatrixLayout(0)
 , /* mM */ 256
 , /* mMmaK */ 32
 , /* mMmaKind */ trtllm::gen::MmaKind(5)
@@ -5634,8 +6734,6 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseShuffledMatrixA */ 1
 , /* mSliceK */ 0
 , /* mSplitK */ gemm::SplitK(0)
-, /* mTransposeMatrixA */ 0
-, /* mTransposeMatrixB */ 1
 , /* mTransposeMmaOutput */ 1
 , /* mTileM */ 128
 , /* mTileN */ 8
@@ -5650,8 +6748,9 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mUseTwoTmaLoadWarps */ 1
 , /* mUseTwoMmaWarps */ 0
 , /* mSfLayoutA */ trtllm::gen::SfLayout(3)
-, /* mSfLayoutB */ trtllm::gen::SfLayout(1)
+, /* mSfLayoutB */ trtllm::gen::SfLayout(0)
 , /* mSfLayoutC */ trtllm::gen::SfLayout(1)
+, /* mSfReshapeFactor */ 1
 , /* mTileScheduler */ gemm::TileScheduler(0)
 , /* mActType */ gemmGatedAct::ActType(0)
 , /* mBatchedM */ {}
@@ -5660,11 +6759,12 @@ static const batchedGemm::BatchedGemmConfig tllmGenBatchedGemmList[] = {
 , /* mNumBatches */ 2
 , /* mIsStaticBatch */ 0
 , /* mNumTokens */ 2
-, /* mRouteAct */ 1
+, /* mRouteImpl */ batchedGemm::RouteImpl(1)
 , /* mGridWaitForPrimaryRouting */ 1
 , /* mFusedAct */ 1
 , /* mNumRegsPerThreadNonEpilogueWarp */ 0
 , /* mNumRegsPerThreadEpilogueWarp */ 0
+, /* mNumRegsCastAWarps */ 0
  }, gemm::SmVersion::Sm100a },
 #endif // EXCLUDE_SM_100
 };
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/KernelParams.h b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/KernelParams.h
index 77dd7e43d7..6fb72cf40d 100644
--- a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/KernelParams.h
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/KernelParams.h
@@ -19,6 +19,7 @@
 #include "trtllm/gen/CommonUtils.h"
 #include "trtllm/gen/SfLayoutDecl.h"
 
+#include "BatchedGemmEnums.h"
 #include "Enums.h"
 #include "TmaDescriptor.h"
 
@@ -65,16 +66,22 @@ struct KernelParams
     //    Tile box strides are [tileK, 1].
     //
     // If batchN:
-    //    If transposeMatrixA is false
+    //    If layoutA is MatrixLayout::MajorK
     //       Logical shape is [B, divUpMul(M, tileM), K].
     //       Logical strides are [divUpMul(M, tileM) * K, K, 1].
     //       Tile box shape is [1, tileM, tileK].
     //       Tile box strides are [0, tileK, 1].
-    //    If transposeMatrixA is true
+    //    If layoutA is MatrixLayout::Mn
     //       Logical shape is [B, K, divUpMul(M, tileM)].
     //       Logical strides are [K * divUpMul(M, tileM), divUpMul(M, tileM), 1].
     //       Tile box shape is [1, tileK, tileM].
     //       Tile box strides are [0, tileM, 1].
+    //    If layoutA is MatrixLayout::BlockMajorK
+    //       Logical shape is [B, K / blockK, divUpMul(M, tileM), blockK].
+    //       Logical strides are [K * divUpMul(M, tileM),  divUpMul(M, tileM) * blockK, blockK, 1].
+    //       Tile box shape is [1, tileK / min(blockK, tileK), tileM, min(blockK, tileK)].
+    //       Tile box strides are [0, tileM * min(blockK, tileK), min(blockK, tileK), 1].
+    //       where blockK is 128B.
     //
     // Dtype is set from options.mDtypeA.
     CUtensorMap tmaA[1];
@@ -84,16 +91,22 @@ struct KernelParams
     // makeTmaShapeStrideAbc.
     //
     // If batchM:
-    //    If transposeMatrixB is true
+    //    If layoutB is MatrixLayout::MajorK
     //       Logical shape is [B, divUpMul(N, tileN), K].
     //       Logical strides are [divUpMul(N, tileN) * K, K, 1].
     //       Tile box shape is [1, tileN, tileK].
     //       Tile box strides are [0, tileK, 1].
-    //    If transposeMatrixB is false
+    //    If layoutB is MatrixLayout::MajorMn
     //       Logical shape is [B, K, divUpMul(N, tileN)].
     //       Logical strides are [K * divUpMul(N, tileN), divUpMul(N, tileN), 1].
     //       Tile box shape is [1, tileK, tileN].
     //       Tile box strides are [0, tileN, 1].
+    //    If layoutB is MatrixLayout::BlockMajorK
+    //       Logical shape is [B, K / blockK, divUpMul(N, tileN), blockK].
+    //       Logical strides are [K * divUpMul(N, tileN),  divUpMul(N, tileN) * blockK, blockK, 1].
+    //       Tile box shape is [1, tileK / min(blockK, tileK), tileN, min(blockK, tileK)].
+    //       Tile box strides are [0, tileN * min(blockK, tileK), min(blockK, tileK), 1].
+    //       where blockK is 128B.
     //
     // If batchN:
     //    Logical shape is [sum(divUpMul(N[bi], tileN) for bi in B), K].
@@ -212,6 +225,11 @@ struct KernelParams
     // Shape is [B]. One scaling factor per tensor in batch.
     float const* ptrScaleGate;
 
+    // The alpha and beta for SwiGlu.
+    // Shape is [B]. One alpha and one beta per tensor in batch.
+    float const* ptrSwiGluAlpha;
+    float const* ptrSwiGluBeta;
+
     // The K dimension. It is the hidden dimension of the input matrices.
     int32_t k;
 
@@ -294,6 +312,21 @@ struct KernelParams
     //     Logical shape is [sum(divUpMul(N[bi], tileN) for bi in B)]
     void const* ptrPerTokenSfB;
 
+    // The bias applied after the GEMM and before the activation function.
+    // The bias is applied before applying the global scaling factor. I.e.
+    // C = act(A * B + bias') * scaleC
+    // scaleC = dequantA * dequantB * quantC
+    // Thus, the bias' = bias / (dequantA * dequantB), where the bias is the original bias.
+    //
+    // If batchM, BiasType must be N, and bias shape is [B, N].
+    // The bias is broadcasted along the M dimension.
+    //
+    // If batchNm BiasType must be M, and bias shape is [B, M].
+    // The bias is broadcasted along the N dimension.
+    //
+    // The dtype is float32.
+    void const* ptrBias{nullptr};
+
     // The output block scaling factors for C.
     //
     // If MxFp{4,8} and NvFp4 formats are used,
@@ -366,6 +399,9 @@ struct KernelParams
     // Total number of unpadded inputs
     int32_t numTokens;
 
+    // Total number of batches
+    int32_t numBatches;
+
     //////////////////////////////////////////////////////////////////////////////////////////////////
     //
     // Batching information parameters.
@@ -478,29 +514,44 @@ struct KernelParams
     };
 
     // Create the TMA shape/stride for A/B/C.
-    static auto makeTmaShapeStrideAbc(bool const transposeMatrixA, bool const transposeMatrixB,
-        bool const transposeMmaOutput, bool const useFusedAct, int numBatches, int mM, int mN, int mK,
-        MatrixType matrixType)
+    template <class GemmOptions>
+    static auto makeTmaShapeStrideAbc(
+        GemmOptions const& options, int mM, int mN, int mK, int tileM, int tileN, int tileK, MatrixType matrixType)
     {
         // Weights matrix is A if we transpose the output of MMA (to have it M-major).
         // Otherwise, it is B, when the output of MMA is K-major.
-        bool const isWeights = (matrixType == MatrixType::MatrixA && transposeMmaOutput)
-            || (matrixType == MatrixType::MatrixB && !transposeMmaOutput);
+        bool const isWeights = (matrixType == MatrixType::MatrixA && options.mTransposeMmaOutput)
+            || (matrixType == MatrixType::MatrixB && !options.mTransposeMmaOutput);
 
+        // The outer dimension.
         auto numTokens = (matrixType == MatrixType::MatrixA || matrixType == MatrixType::MatrixC) ? mM : mN;
+        // The outer dimension tile size.
+        auto tileNumTokens = (matrixType == MatrixType::MatrixC) ? options.mEpilogueTileM
+            : (matrixType == MatrixType::MatrixA)                ? tileM
+                                                                 : tileN;
+        // The inner dimension.
         auto hiddenSize = (matrixType == MatrixType::MatrixC) ? mN : mK;
-        if (matrixType == MatrixType::MatrixC && transposeMmaOutput)
+        // The inner dimension tile size.
+        auto tileHiddenSize = (matrixType == MatrixType::MatrixC) ? options.mEpilogueTileN : tileK;
+
+        // Swap matrix C sizes if output is transpose
+        if (matrixType == MatrixType::MatrixC && options.mTransposeMmaOutput)
         {
             numTokens = mN;
             hiddenSize = mM;
+            tileNumTokens = options.mEpilogueTileN;
+            tileHiddenSize = options.mEpilogueTileM;
         }
 
         // For a fused activation kernel, the hidden size of output is halved. TODO: That's true for
         // gated activations but not regular activations.
-        if (useFusedAct)
+        if (options.mFusedAct)
         {
             if (matrixType == MatrixType::MatrixC)
+            {
                 hiddenSize /= 2;
+                tileHiddenSize /= 2;
+            }
         }
 
         // The cute tensor shape for A/B: (numTokens, hiddenSize).
@@ -511,7 +562,7 @@ struct KernelParams
         // Ativations matrix is 2D (sum(divUpMul(M[bi], tileM) for bi in B), K).
         if (isWeights)
         {
-            shape.push_back(static_cast<uint64_t>(numBatches));
+            shape.push_back(static_cast<uint64_t>(options.mNumBatches));
         }
 
         // Assemble the stride (strideTokens, 1).
@@ -522,20 +573,41 @@ struct KernelParams
             stride.push_back(static_cast<uint64_t>(hiddenSize * numTokens));
         }
 
-        // Apply transpose if necessary
-        if ((matrixType == MatrixType::MatrixA && transposeMatrixA)
-            || (matrixType == MatrixType::MatrixB && !transposeMatrixB))
+        // Assemble the box shape
+        std::vector<int32_t> tileShape = {tileHiddenSize, tileNumTokens};
+
+        // Alternate layouts do not apply to matrixC
+        if (matrixType != MatrixType::MatrixC)
         {
-            std::swap(shape[0], shape[1]);
-            stride[1] = numTokens;
+            gemm::MatrixLayout layout = (matrixType == MatrixType::MatrixA) ? options.mLayoutA : options.mLayoutB;
+            // Note, only the weights support non MajorK layouts
+            if (layout == gemm::MatrixLayout::MajorMn)
+            {
+                // Apply transpose if necessary
+                std::swap(shape[0], shape[1]);
+                stride[1] = numTokens;
+                std::swap(tileShape[0], tileShape[1]);
+            }
+            else if (layout == gemm::MatrixLayout::BlockMajorK)
+            {
+                // Set shapes based on blocking layout
+                shape = {static_cast<uint64_t>(options.mBlockK), static_cast<uint64_t>(numTokens),
+                    static_cast<uint64_t>(mK / options.mBlockK), static_cast<uint64_t>(options.mNumBatches)};
+                stride = {1, static_cast<uint64_t>(options.mBlockK), static_cast<uint64_t>(numTokens * options.mBlockK),
+                    static_cast<uint64_t>(hiddenSize * numTokens)};
+
+                // If blockK > tileK, then the inner most box size will be based on the tile
+                int32_t const tileBlockK = std::min(options.mBlockK, tileHiddenSize);
+                tileShape = {tileBlockK, tileNumTokens, tileHiddenSize / tileBlockK};
+            }
         }
 
-        return std::make_tuple(shape, stride);
+        return std::make_tuple(shape, stride, tileShape);
     }
 
     // Create the TMA shape/stride for A/B block scaling factors.
     static auto makeTmaShapeStrideSfAb(int mM, int mN, int mK, MatrixType matrixType, int tileM, int tileN, int tileK,
-        tg::Dtype dtypeElt, tg::SfLayout layout)
+        tg::Dtype dtypeElt, tg::SfLayout layout, int sfReshapeFactor)
     {
 
         // The outer dimension.
@@ -585,15 +657,36 @@ struct KernelParams
         {
             // The scaling factor tensor packs 8x4 tiles into contiguous 32B blocks.
             //
-            // As the inner dimension (k) is required to be a multiple of the tile size, we
-            // can reshape to use fewer read requests, if the tile dimensions allow.
-            // I.e., let's define repeats = min(hiddenSizePerTile / numEltsPerSf / 4, 8)
+            // As the inner dimension (k) is often a multiple of the tile size, we can reshape to use
+            // fewer read requests, if the tile dimensions allow. It does not reduce the number of
+            // instructions.
+            //
+            // I.e., let's define r = min(⌈hiddenSizePerTile / (numEltsPerSf * 4)⌉, 8)
+            //
+            // The "logical" tensor is: [outer,      inner / numEltsPerSf]
+            // The 8x4 SF layout is:    [⌈outer / 8⌉, inner / (4 * numEltsPerSf), 32]
+            // The TMA tensor shape is: [⌈outer / 8⌉, inner / (4 * numEltsPerSf * r), r * 32]
             //
-            // The "logical" tensor is: [outer,     inner / numEltsPerSf]
-            // The 8x4 SF layout is:    [outer / 8, inner / numEltsPerSf / 4, 32]
-            // The TMA tensor shape is: [outer / 8, inner / numEltsPerSf / 4 / repeats, repeats * 32]
+            // The caveat of NumRepeats>1 is we must pad the hidden dimension of SF to multiples of
+            // NumRepeats * numEltsPerSf * 4.
+
+            // Detect if the supplied factor is power of 2. E.g., 0b0100 and (0b0100 - 1) == 0b0000.
+            int const r = sfReshapeFactor;
+            if (r > 0 && (r & (r - 1)) != 0)
+            {
+                throw std::runtime_error(
+                    "mSfReshapeFactor must be positive and a power of 2. Found " + std::to_string(r));
+            }
 
-            int const repeats = std::min(ceilDiv(hiddenSizePerTile, numEltsPerSf * 4), 8);
+            // Sanitize number of repeats so it doesn't exceed the dimension.
+            int const repeats = std::min(ceilDiv(hiddenSizePerTile, numEltsPerSf * 4), r);
+
+            // Detect if the input hidden size K is a multiple of the repeats.
+            if (ceilDiv(hiddenSize, numEltsPerSf * 4) % repeats != 0)
+            {
+                throw std::runtime_error("SF hiddenSize K (" + std::to_string(ceilDiv(hiddenSize, numEltsPerSf * 4))
+                    + ") must be a multiple of repeats (" + std::to_string(repeats) + ")");
+            }
 
             auto shape = std::vector<uint64_t>{static_cast<uint64_t>(repeats * 32),
                 static_cast<uint64_t>(ceilDiv(hiddenSize, numEltsPerSf * 4 * repeats)),
@@ -618,18 +711,14 @@ struct KernelParams
         return std::make_tuple(std::vector<uint64_t>{}, std::vector<uint64_t>{}, std::vector<uint32_t>{});
     }
 
-    static KernelParams setKernelParams(int32_t const numBatches, int32_t const numTokens, bool const batchM,
-        int32_t const m, int32_t const n, int32_t const k, std::vector<int32_t> const& batchedM,
-        std::vector<int32_t> const& batchedN, int32_t const tileM, int32_t const tileN, int32_t const tileK,
-        int32_t const epilogueTileM, int32_t const epilogueTileN, bool const useDeepSeekFp8, bool const useTmaStore,
-        bool const transposeMatrixA, bool const transposeMatrixB, bool const transposeMmaOutput, tg::SfLayout sfLayoutB,
-        bool const useFusedAct, tg::Dtype dtypeA, tg::Dtype dtypeB, tg::Dtype dtypeC, tg::MmaKind mmaKind,
-        void const* ptrA, void const* ptrB, void* ptrC, void const* dSfA, void const* dSfB, void const* ptrPerTokenSfA,
-        void const* ptrPerTokenSfB, void* dSfC, float const* ptrScaleC, float const* ptrScaleGate,
-        int32_t const* ptrRouteMap, float* rowMax, uint32_t* rowMaxBars, bool isStaticBatch = true,
-        int32_t const* ptrNumNonExitingCtas = nullptr, int32_t const* ptrTotalNumPaddedTokens = nullptr,
-        int32_t const* ptrCtaIdxXyToBatchIdx = nullptr, int32_t const* ptrCtaIdxXyToMnLimit = nullptr,
-        int32_t const maxNumCtas = MaxNumCtas)
+    template <class GemmOptions_>
+    static KernelParams setKernelParams(GemmOptions_ const& options, bool const batchM, void const* ptrA,
+        void const* ptrB, void* ptrC, void const* dSfA, void const* dSfB, void const* ptrPerTokenSfA,
+        void const* ptrPerTokenSfB, void const* ptrBias, void* dSfC, float const* ptrScaleC, float const* ptrScaleGate,
+        float const* ptrSwiGluAlpha, float const* ptrSwiGluBeta, int32_t const* routeMap, float* rowMax,
+        uint32_t* rowMaxBars, int32_t const* ptrNumNonExitingCtas = nullptr,
+        int32_t const* ptrTotalNumPaddedTokens = nullptr, int32_t const* ptrCtaIdxXyToBatchIdx = nullptr,
+        int32_t const* ptrCtaIdxXyToMnLimit = nullptr, int32_t const maxNumCtas = MaxNumCtas)
     {
 
         static_assert(sizeof(KernelParams) <= 32 * 1024, "sizeof(KernelParams) has to be less or equal than 32KB");
@@ -637,31 +726,34 @@ struct KernelParams
         // Create the return struct.
         KernelParams params;
 
-        if (numBatches > KernelParams::MaxBatchSize)
+        if (options.mNumBatches > KernelParams::MaxBatchSize)
         {
             throw std::runtime_error("GEMM batch limit reached.");
         }
 
-        params.ptrRouteMap = ptrRouteMap;
-        params.numTokens = numTokens;
+        params.ptrRouteMap = routeMap;
+        params.numTokens = options.mNumTokens;
 
         params.ptrScaleC = ptrScaleC;
         params.ptrScaleGate = ptrScaleGate;
 
+        params.ptrSwiGluAlpha = ptrSwiGluAlpha;
+        params.ptrSwiGluBeta = ptrSwiGluBeta;
+
         int32_t ctaOffset = 0;
 
         // Compute totalNumPaddedTokens, ctaIdxXyToBatchIdx and ctaIdxXyToMnLimit if the batch dims are
         // known at kernel launch time. Otherwise, these parameters are defined in the device buffers:
         // ptrTotalNumPaddedTokens, ptrCtaIdxXyToBatchIdx and ptrCtaIdxXyToMnLimit respectively.
 
-        if (isStaticBatch)
+        if (options.mIsStaticBatch)
         {
             params.totalNumPaddedTokens = 0;
-            for (int b = 0; b < numBatches; b++)
+            for (int b = 0; b < options.mNumBatches; b++)
             {
 
-                int mM = batchM ? batchedM[b] : n;
-                int mN = batchM ? m : batchedN[b];
+                int mM = batchM ? options.mBatchedM[b] : options.mN;
+                int mN = batchM ? options.mM : options.mBatchedN[b];
 
                 // Skip Tma descriptor creation if expert isn't used
                 if (mM == 0 || mN == 0)
@@ -670,9 +762,10 @@ struct KernelParams
                 }
 
                 // The number of CTAs.
-                int32_t numCtas = batchM ? (mM + tileM - 1) / tileM : (mN + tileN - 1) / tileN;
+                int32_t numCtas
+                    = batchM ? (mM + options.mTileM - 1) / options.mTileM : (mN + options.mTileN - 1) / options.mTileN;
                 // The size of the tile.
-                int32_t tile = batchM ? tileM : tileN;
+                int32_t tile = batchM ? options.mTileM : options.mTileN;
                 // The problem size.
                 int32_t mn = batchM ? mM : mN;
                 int32_t tokensPerTile = mn;
@@ -704,15 +797,15 @@ struct KernelParams
             ctaOffset = maxNumCtas;
         }
 
-        if (useDeepSeekFp8 && dtypeC == tg::Dtype::E4m3)
+        if (options.mUseDeepSeekFp8 && options.mDtypeC == tg::Dtype::E4m3)
         {
             params.ptrDqSfsC = reinterpret_cast<float*>(dSfC);
         }
 
         params.ptrA = ptrA;
         params.ptrB = ptrB;
-        params.strideInBytesA = k * tg::dtypeGetNumBits(dtypeA) / 8;
-        params.strideInBytesB = k * tg::dtypeGetNumBits(dtypeB) / 8;
+        params.strideInBytesA = options.mK * tg::dtypeGetNumBits(options.mDtypeA) / 8;
+        params.strideInBytesB = options.mK * tg::dtypeGetNumBits(options.mDtypeB) / 8;
 
         params.ptrSfA = dSfA;
         params.ptrSfB = dSfB;
@@ -721,75 +814,96 @@ struct KernelParams
         if (!batchM)
         {
             // A is the expert
-            if (0 != m % tileM)
+            if (0 != options.mM % options.mTileM)
             {
                 throw std::runtime_error("0 == mM %% tileM");
             }
-            params.tileStridePerBatch = m / tileM;
-            params.nm = m;
+            params.tileStridePerBatch = options.mM / options.mTileM;
+            params.nm = options.mM;
             // Shape/stride for gmem tensor A.
-            auto [shapeA, strideA] = makeTmaShapeStrideAbc(transposeMatrixA, transposeMatrixB, transposeMmaOutput,
-                useFusedAct, numBatches, m, n, k, MatrixType::MatrixA);
+            auto [shapeA, strideA, tileShapeA] = makeTmaShapeStrideAbc(options, options.mM, options.mN, options.mK,
+                options.mTileM, options.mTileN, options.mTileK, MatrixType::MatrixA);
             // Build tma descriptor for A.
-            params.tmaA[0] = gemm::buildNdTmaDescriptor(dtypeA, mmaKind, shapeA, strideA,
-                transposeMatrixA ? tileK : tileM, transposeMatrixA ? tileM : tileK, const_cast<void*>(ptrA));
+            params.tmaA[0] = gemm::buildNdTmaDescriptor(
+                options.mDtypeA, options.mMmaKind, shapeA, strideA, tileShapeA, const_cast<void*>(ptrA));
 
             // The input is padded:
             // [act0, padding, padding, ... TileN size .., act1, padding, padding, ...]
-            auto const inputNumTokens = ctaOffset * tileN;
-            // B is the activation
-            // Shape/stride for gmem tensor B.
-            auto [shapeB, strideB] = makeTmaShapeStrideAbc(transposeMatrixA, transposeMatrixB, transposeMmaOutput,
-                useFusedAct, numBatches, m, inputNumTokens, k, MatrixType::MatrixB);
-            // Build tma descriptor for B.
-            params.tmaB[0] = gemm::buildNdTmaDescriptor(dtypeB, mmaKind, shapeB, strideB,
-                !transposeMatrixB ? tileK : tileN, !transposeMatrixB ? tileN : tileK, const_cast<void*>(ptrB));
+            auto const inputNumTokens = ctaOffset * options.mTileN;
+
+            if (!batchedGemm::doesRouteImplUseLdgsts(options.mRouteImpl))
+            {
+                bool useRouteAct = batchedGemm::doesRouteImplUseTma(options.mRouteImpl);
+                // B is the activation
+                // Shape/stride for gmem tensor B.
+                auto [shapeB, strideB, tileShapeB] = makeTmaShapeStrideAbc(options, options.mM,
+                    useRouteAct ? options.mNumTokens : inputNumTokens, options.mK, options.mTileM,
+                    (useRouteAct ? 1 : options.mTileN), options.mTileK, MatrixType::MatrixB);
+                // Build tma descriptor for B.
+                params.tmaB[0] = gemm::buildNdTmaDescriptor(
+                    options.mDtypeB, options.mMmaKind, shapeB, strideB, tileShapeB, const_cast<void*>(ptrB));
+            }
 
-            if (dtypeA == tg::Dtype::E2m1 || dtypeA == tg::Dtype::MxE4m3 || dtypeA == tg::Dtype::MxE2m1)
+            if (options.mDtypeA == tg::Dtype::E2m1 || options.mDtypeA == tg::Dtype::MxE4m3
+                || options.mDtypeA == tg::Dtype::MxE2m1)
             {
-                tg::Dtype const dTypeSf = (dtypeA == tg::Dtype::E2m1) ? tg::Dtype::E4m3 : tg::Dtype::UE8m0;
+                tg::Dtype const dTypeSf = (options.mDtypeA == tg::Dtype::E2m1) ? tg::Dtype::E4m3 : tg::Dtype::UE8m0;
 
                 // Build TMA descriptor for gmem A block scaling factors.
-                auto [shapeSfA, strideSfA, tileShapesSfA] = makeTmaShapeStrideSfAb(
-                    m * numBatches, n, k, MatrixType::MatrixA, tileM, tileN, tileK, dtypeA, tg::SfLayout::R128c4);
+                auto [shapeSfA, strideSfA, tileShapesSfA] = makeTmaShapeStrideSfAb(options.mM * options.mNumBatches,
+                    options.mN, options.mK, MatrixType::MatrixA, options.mTileM, options.mTileN, options.mTileK,
+                    options.mDtypeA, tg::SfLayout::R128c4, options.mSfReshapeFactor);
                 params.tmaSfA[0]
                     = gemm::buildSfTmaDescriptor(dTypeSf, shapeSfA, strideSfA, tileShapesSfA, const_cast<void*>(dSfA));
             }
 
-            if (dtypeB == tg::Dtype::E2m1 || dtypeB == tg::Dtype::MxE4m3 || dtypeB == tg::Dtype::MxE2m1)
+            if (options.mDtypeB == tg::Dtype::E2m1 || options.mDtypeB == tg::Dtype::MxE4m3
+                || options.mDtypeB == tg::Dtype::MxE2m1)
             {
-                tg::Dtype const dTypeSf = (dtypeB == tg::Dtype::E2m1) ? tg::Dtype::E4m3 : tg::Dtype::UE8m0;
+                tg::Dtype const dTypeSf = (options.mDtypeB == tg::Dtype::E2m1) ? tg::Dtype::E4m3 : tg::Dtype::UE8m0;
 
-                // The input is padded:
-                // [act0, padding, padding, ... TileN size .., act1, padding, padding, ...]
-                auto const inputNumTokensSfB = ctaOffset * tileN;
+                if (batchedGemm::doesRouteImplUseTma(options.mRouteImpl))
+                {
 
-                // Build TMA descriptor for gmem B block scaling factors.
-                auto [shapeSfB, strideSfB, tileShapesSfB] = makeTmaShapeStrideSfAb(
-                    m, inputNumTokensSfB, k, MatrixType::MatrixB, tileM, tileN, tileK, dtypeB, sfLayoutB);
-                params.tmaSfB[0]
-                    = gemm::buildSfTmaDescriptor(dTypeSf, shapeSfB, strideSfB, tileShapesSfB, const_cast<void*>(dSfB));
+                    // The input is NOT padded:
+                    // [act0, act1, act2, ...]
+
+                    // Build TMA descriptor for gmem B block scaling factors.
+                    int32_t const numEltsPerSf = tg::dtypeNumEltsPerSf(options.mDtypeB);
+                    auto [shapeSfB, strideSfB, tileShapesSfB]
+                        = makeTmaShapeStrideAbc(options, options.mM, options.mNumTokens, options.mK / numEltsPerSf,
+                            options.mTileM, 1 /* tileN */, options.mTileK / numEltsPerSf, MatrixType::MatrixB);
+                    params.tmaSfB[0] = gemm::buildNdTmaDescriptor(dTypeSf, options.mMmaKind, shapeSfB, strideSfB,
+                        tileShapesSfB, const_cast<void*>(dSfB),
+                        /*doSwizzle*/ true);
+                }
+                else if (batchedGemm::doesRouteImplUseNoRoute(options.mRouteImpl))
+                {
+
+                    // The input is padded:
+                    // [act0, padding, padding, ... TileN size .., act1, padding, padding, ...]
+
+                    auto const inputNumTokensSfB = ctaOffset * options.mTileN;
+
+                    // Build TMA descriptor for gmem B block scaling factors.
+                    auto [shapeSfB, strideSfB, tileShapesSfB] = makeTmaShapeStrideSfAb(options.mM, inputNumTokensSfB,
+                        options.mK, MatrixType::MatrixB, options.mTileM, options.mTileN, options.mTileK,
+                        options.mDtypeB, options.mSfLayoutB, options.mSfReshapeFactor);
+                    params.tmaSfB[0] = gemm::buildSfTmaDescriptor(
+                        dTypeSf, shapeSfB, strideSfB, tileShapesSfB, const_cast<void*>(dSfB));
+                }
             }
 
             // C is the output activation
-            if (useTmaStore)
+            if (options.mUseTmaStore)
             {
                 // Shape/stride for gmem tensor C.
-                auto [shapeC, strideC] = makeTmaShapeStrideAbc(transposeMatrixA, transposeMatrixB, transposeMmaOutput,
-                    useFusedAct, numBatches, m, ctaOffset * tileN, k, MatrixType::MatrixC);
-
-                // Swap M and N tiles for the M-major epilogue.
-                auto outputTileM = transposeMmaOutput ? epilogueTileN : epilogueTileM;
-                auto outputTileN = transposeMmaOutput ? epilogueTileM : epilogueTileN;
-
-                if (useFusedAct)
-                {
-                    // for a fused activation kernel, output tile `N` is halved
-                    outputTileN /= 2;
-                }
+                auto [shapeC, strideC, tileShapeC]
+                    = makeTmaShapeStrideAbc(options, options.mM, ctaOffset * options.mTileN, options.mK, options.mTileM,
+                        options.mTileN, options.mTileK, MatrixType::MatrixC);
                 // Build tma descriptor for C.
-                params.tmaC[0] = gemm::buildNdTmaDescriptor(
-                    dtypeC, tg::MmaKind::Auto, shapeC, strideC, outputTileM, outputTileN, ptrC);
+                params.tmaC[0]
+                    = gemm::buildNdTmaDescriptor(options.mDtypeC, tg::MmaKind::Auto, shapeC, strideC, tileShapeC, ptrC);
             }
             else
             {
@@ -799,75 +913,76 @@ struct KernelParams
         else
         {
             // B is the expert
-            if (0 != n % tileN)
+            if (0 != options.mN % options.mTileN)
             {
                 throw std::runtime_error("0 == mN %% tileN");
             }
-            params.tileStridePerBatch = n / tileN;
-            params.nm = n;
+            params.tileStridePerBatch = options.mN / options.mTileN;
+            params.nm = options.mN;
             // Shape/stride for gmem tensor B.
-            auto [shapeB, strideB] = makeTmaShapeStrideAbc(transposeMatrixA, transposeMatrixB, transposeMmaOutput,
-                useFusedAct, numBatches, m, n, k, MatrixType::MatrixB);
+            auto [shapeB, strideB, tileShapeB] = makeTmaShapeStrideAbc(options, options.mM, options.mN, options.mK,
+                options.mTileM, options.mTileN, options.mTileK, MatrixType::MatrixB);
             // Build tma descriptor for B.
-            params.tmaB[0] = gemm::buildNdTmaDescriptor(dtypeB, mmaKind, shapeB, strideB,
-                !transposeMatrixB ? tileK : tileN, !transposeMatrixB ? tileN : tileK, const_cast<void*>(ptrB));
+            params.tmaB[0] = gemm::buildNdTmaDescriptor(
+                options.mDtypeB, options.mMmaKind, shapeB, strideB, tileShapeB, const_cast<void*>(ptrB));
 
-            // A is the activation
-            // Shape/stride for gmem tensor A.
-            // The input is padded:
-            // [act0, padding, padding, ... tileM size .., act1, padding, padding, ...]
-            auto const inputNumTokens = ctaOffset * tileM;
-            auto [shapeA, strideA] = makeTmaShapeStrideAbc(transposeMatrixA, transposeMatrixB, transposeMmaOutput,
-                useFusedAct, numBatches, inputNumTokens, n, k, MatrixType::MatrixA);
-            // Build tma descriptor for A.
-            params.tmaA[0] = gemm::buildNdTmaDescriptor(dtypeA, mmaKind, shapeA, strideA,
-                transposeMatrixA ? tileK : tileM, transposeMatrixA ? tileM : tileK, const_cast<void*>(ptrA));
-
-            if (dtypeA == tg::Dtype::E2m1 || dtypeA == tg::Dtype::MxE4m3 || dtypeA == tg::Dtype::MxE2m1)
+            if (options.mRouteImpl == batchedGemm::RouteImpl::NoRoute)
             {
-                tg::Dtype const dTypeSf = (dtypeA == tg::Dtype::E2m1) ? tg::Dtype::E4m3 : tg::Dtype::UE8m0;
-
+                // A is the activation
+                // Shape/stride for gmem tensor A.
                 // The input is padded:
                 // [act0, padding, padding, ... tileM size .., act1, padding, padding, ...]
-                auto const inputNumTokensSfA = ctaOffset * tileM;
+                auto const inputNumTokens = ctaOffset * options.mTileM;
+                auto [shapeA, strideA, tileShapeA] = makeTmaShapeStrideAbc(options, inputNumTokens, options.mN,
+                    options.mK, options.mTileM, options.mTileN, options.mTileK, MatrixType::MatrixA);
+                // Build tma descriptor for A.
+                params.tmaA[0] = gemm::buildNdTmaDescriptor(
+                    options.mDtypeA, options.mMmaKind, shapeA, strideA, tileShapeA, const_cast<void*>(ptrA));
+            }
 
-                // Build TMA descriptor for gmem A block scaling factors.
-                auto [shapeSfA, strideSfA, tileShapesSfA] = makeTmaShapeStrideSfAb(
-                    inputNumTokensSfA, n, k, MatrixType::MatrixA, tileM, tileN, tileK, dtypeA, tg::SfLayout::R128c4);
-                params.tmaSfA[0]
-                    = gemm::buildSfTmaDescriptor(dTypeSf, shapeSfA, strideSfA, tileShapesSfA, const_cast<void*>(dSfA));
+            if (options.mDtypeA == tg::Dtype::E2m1 || options.mDtypeA == tg::Dtype::MxE4m3
+                || options.mDtypeA == tg::Dtype::MxE2m1)
+            {
+                tg::Dtype const dTypeSf = (options.mDtypeA == tg::Dtype::E2m1) ? tg::Dtype::E4m3 : tg::Dtype::UE8m0;
+
+                if (options.mRouteImpl == batchedGemm::RouteImpl::NoRoute)
+                {
+
+                    // The input is padded:
+                    // [act0, padding, padding, ... tileM size .., act1, padding, padding, ...]
+                    auto const inputNumTokensSfA = ctaOffset * options.mTileM;
+
+                    // Build TMA descriptor for gmem A block scaling factors.
+                    auto [shapeSfA, strideSfA, tileShapesSfA] = makeTmaShapeStrideSfAb(inputNumTokensSfA, options.mN,
+                        options.mK, MatrixType::MatrixA, options.mTileM, options.mTileN, options.mTileK,
+                        options.mDtypeA, tg::SfLayout::R128c4, options.mSfReshapeFactor);
+                    params.tmaSfA[0] = gemm::buildSfTmaDescriptor(
+                        dTypeSf, shapeSfA, strideSfA, tileShapesSfA, const_cast<void*>(dSfA));
+                }
             }
 
-            if (dtypeB == tg::Dtype::E2m1 || dtypeB == tg::Dtype::MxE4m3 || dtypeB == tg::Dtype::MxE2m1)
+            if (options.mDtypeB == tg::Dtype::E2m1 || options.mDtypeB == tg::Dtype::MxE4m3
+                || options.mDtypeB == tg::Dtype::MxE2m1)
             {
-                tg::Dtype const dTypeSf = (dtypeB == tg::Dtype::E2m1) ? tg::Dtype::E4m3 : tg::Dtype::UE8m0;
+                tg::Dtype const dTypeSf = (options.mDtypeB == tg::Dtype::E2m1) ? tg::Dtype::E4m3 : tg::Dtype::UE8m0;
 
                 // Build TMA descriptor for gmem B block scaling factors.
-                auto [shapeSfB, strideSfB, tileShapesSfB] = makeTmaShapeStrideSfAb(
-                    m, n * numBatches, k, MatrixType::MatrixB, tileM, tileN, tileK, dtypeB, sfLayoutB);
+                auto [shapeSfB, strideSfB, tileShapesSfB] = makeTmaShapeStrideSfAb(options.mM,
+                    options.mN * options.mNumBatches, options.mK, MatrixType::MatrixB, options.mTileM, options.mTileN,
+                    options.mTileK, options.mDtypeB, options.mSfLayoutB, options.mSfReshapeFactor);
                 params.tmaSfB[0]
                     = gemm::buildSfTmaDescriptor(dTypeSf, shapeSfB, strideSfB, tileShapesSfB, const_cast<void*>(dSfB));
             }
 
             // C is the output activation
-            if (useTmaStore)
+            if (options.mUseTmaStore)
             {
                 // Shape/stride for gmem tensor C.
-                auto [shapeC, strideC] = makeTmaShapeStrideAbc(transposeMatrixA, transposeMatrixB, transposeMmaOutput,
-                    useFusedAct, numBatches, ctaOffset * tileM, n, k, MatrixType::MatrixC);
-
-                // Swap M and N tiles for the M-major epilogue.
-                auto outputTileM = transposeMmaOutput ? epilogueTileN : epilogueTileM;
-                auto outputTileN = transposeMmaOutput ? epilogueTileM : epilogueTileN;
-
-                if (useFusedAct)
-                {
-                    // for a fused activation kernel, output tile `N` is halved
-                    outputTileN /= 2;
-                }
+                auto [shapeC, strideC, tileShapeC] = makeTmaShapeStrideAbc(options, ctaOffset * options.mTileM,
+                    options.mN, options.mK, options.mTileM, options.mTileN, options.mTileK, MatrixType::MatrixC);
                 // Build tma descriptor for C.
-                params.tmaC[0] = gemm::buildNdTmaDescriptor(
-                    dtypeC, tg::MmaKind::Auto, shapeC, strideC, outputTileM, outputTileN, ptrC);
+                params.tmaC[0]
+                    = gemm::buildNdTmaDescriptor(options.mDtypeC, tg::MmaKind::Auto, shapeC, strideC, tileShapeC, ptrC);
             }
             else
             {
@@ -875,7 +990,8 @@ struct KernelParams
             }
         }
 
-        params.k = k;
+        params.k = options.mK;
+        params.numBatches = options.mNumBatches;
 
         params.rank = 0;
         params.tpGrpSize = 1;
@@ -888,32 +1004,10 @@ struct KernelParams
         // Set the per-token scale factors for MetaFP8 or scale inputs
         params.ptrPerTokenSfA = ptrPerTokenSfA;
         params.ptrPerTokenSfB = ptrPerTokenSfB;
+        params.ptrBias = ptrBias;
 
         return params;
     }
-
-    ////////////////////////////////////////////////////////////////////////////////////////////////////
-
-    template <class GemmOptions_>
-    static KernelParams setKernelParams(GemmOptions_ const& options, bool const batchM, void const* ptrA,
-        void const* ptrB, void* ptrC, void const* dSfA, void const* dSfB, void const* ptrPerTokenSfA,
-        void const* ptrPerTokenSfB, void* dSfC, float const* ptrScaleC, float const* ptrScaleGate,
-        int32_t const* routeMap, float* rowMax, uint32_t* rowMaxBars, int32_t const* ptrNumNonExitingCtas = nullptr,
-        int32_t const* ptrTotalNumPaddedTokens = nullptr, int32_t const* ptrCtaIdxXyToBatchIdx = nullptr,
-        int32_t const* ptrCtaIdxXyToMnLimit = nullptr, int32_t const maxNumCtas = MaxNumCtas)
-    {
-
-        bool const useFusedAct = options.mFusedAct;
-
-        return setKernelParams(options.mNumBatches, options.mNumTokens, batchM, options.mM, options.mN, options.mK,
-            options.mBatchedM, options.mBatchedN, options.mTileM, options.mTileN, options.mTileK,
-            options.mEpilogueTileM, options.mEpilogueTileN, options.mUseDeepSeekFp8, options.mUseTmaStore,
-            options.mTransposeMatrixA, options.mTransposeMatrixB, options.mTransposeMmaOutput, options.mSfLayoutB,
-            useFusedAct, options.mDtypeA, options.mDtypeB, options.mDtypeC, options.mMmaKind, ptrA, ptrB, ptrC, dSfA,
-            dSfB, ptrPerTokenSfA, ptrPerTokenSfB, dSfC, ptrScaleC, ptrScaleGate, routeMap, rowMax, rowMaxBars,
-            options.mIsStaticBatch, ptrNumNonExitingCtas, ptrTotalNumPaddedTokens, ptrCtaIdxXyToBatchIdx,
-            ptrCtaIdxXyToMnLimit, maxNumCtas);
-    }
 #endif
 };
 
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/KernelTraits.h b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/KernelTraits.h
index c46578e4b4..73ce5ce1ab 100644
--- a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/KernelTraits.h
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/KernelTraits.h
@@ -162,11 +162,12 @@ class KernelTraits
     KernelTraits() {}
 
     // The constructor.
-    KernelTraits(tg::Dtype dtypeA, tg::Dtype dtypeB, tg::Dtype dtypeC, tg::Dtype dtypeAcc, tg::MmaKind mmaKind,
-        int32_t tileM, int32_t tileN, int32_t tileK, int32_t epilogueTileM, int32_t epilogueTileN, int32_t numStages,
-        int32_t numStagesMma, int32_t numSlicesForSplitK, int32_t numSlicesForSliceK, SplitK splitK, bool useTmaStore,
-        bool transposeMmaOutput, AllReduceAlgo allReduceAlgo, bool usePersistentScheduler, bool useDeepSeekFp8,
-        bool usePerTokenSfA, bool usePerTokenSfB)
+    KernelTraits(tg::Dtype dtypeA, tg::Dtype dtypeB, tg::Dtype dtypeC, tg::Dtype dtypeAcc, tg::Dtype dtypeMmaA,
+        tg::Dtype dtypeMmaB, tg::MmaKind mmaKind, int32_t tileM, int32_t tileN, int32_t tileK, int32_t epilogueTileM,
+        int32_t epilogueTileN, int32_t numStages, int32_t numStagesMma, int32_t numSlicesForSplitK,
+        int32_t numSlicesForSliceK, SplitK splitK, bool useTmaStore, bool transposeMmaOutput,
+        AllReduceAlgo allReduceAlgo, bool usePersistentScheduler, bool useDeepSeekFp8, bool usePerTokenSfA,
+        bool usePerTokenSfB, BiasType biasType)
         : mMmaKind{mmaKind}
     {
         //
@@ -181,16 +182,17 @@ class KernelTraits
             // [rowMax       ] (16B aligned) (if needed)
             // [sliceK       ] (16B aligned) (if needed)
             // [per-token SF ] (16B aligned) (if needed)
+            // [bias         ] (16B aligned) (if needed)
             //
             // SMEM for smemA and smemB might be repurposed and used for gmemC0 and gmemC1:
             //
             // [..smemA..][..smemB..][..smemBShuffle..]
-            // [..gmemC0..][..gmemC1..][..rowMax..][..sliceK..]
+            // [..gmemC0..][..gmemC1..][..rowMax..][..sliceK..][..per-token SF..][..bias..]
             //
 
             if (mMmaKind == tg::MmaKind::Auto)
             {
-                mMmaKind = dtypeGetMmaKind(dtypeA, dtypeB);
+                mMmaKind = dtypeGetMmaKind(dtypeMmaA, dtypeMmaB);
             }
 
             std::vector<std::pair<int32_t, int32_t>> numBytesAndAlignmentPerSmemChunk;
@@ -344,6 +346,29 @@ class KernelTraits
                 firstChunkReuseSmem.emplace_back(false);
             }
 
+            // Bias
+            {
+                int32_t numBytesSmemBias = 0;
+                if (isBiasTypeN(biasType))
+                {
+                    numBytesSmemBias = tileN * sizeof(float);
+                }
+                else if (isBiasTypeM(biasType))
+                {
+                    numBytesSmemBias = tileM * sizeof(float);
+                }
+                else if (isBiasTypeMn(biasType))
+                {
+                    numBytesSmemBias = tileM * tileN * sizeof(float);
+                }
+                // Number of bytes alignment for bias
+                auto const numBytesAlignmentBias = 16;
+                // Add info.
+                smemChunkNames.emplace_back("smemBias");
+                numBytesAndAlignmentPerSmemChunk.emplace_back(std::make_pair(numBytesSmemBias, numBytesAlignmentBias));
+                firstChunkReuseSmem.emplace_back(false);
+            }
+
             // Per-block absolute maximum for multi-warp reduction.
             {
                 // Number of bytes: number of epilogue warps * number of tile columns.
@@ -401,10 +426,12 @@ class KernelTraits
 
             // Matrix A
             {
+                // We use TMEM for A if we use slice-K or if we need to cast A.
+                bool const useTmemA = (numSlicesForSliceK > 1) || (dtypeMmaA != dtypeA);
                 // Number of columns for A.
-                auto const numTmemColsA = numSlicesForSliceK > 1 ? numStages * tileK
-                        / (numSlicesForSliceK * tg::dtypeGetNumBits(tg::Dtype::UInt32) / tg::dtypeGetNumBits(dtypeA))
-                                                                 : 0;
+                auto const numTmemColsA = useTmemA ? numStages * tileK
+                        / (numSlicesForSliceK * tg::dtypeGetNumBits(tg::Dtype::UInt32) / tg::dtypeGetNumBits(dtypeMmaA))
+                                                   : 0;
                 // Number of columns for A alignment.
                 auto const numColsAlignmentA = 4;
                 // No need to reuse TMEM.
@@ -418,7 +445,7 @@ class KernelTraits
 
             // Sf A
             {
-                bool const useBlockScalingA = tg::dtypeIsBlockFmt(dtypeA);
+                bool const useBlockScalingA = tg::dtypeIsBlockFmt(dtypeMmaA);
                 // Number of columns for scaling factors of A.
                 auto const numTmemColsSfA
                     = useBlockScalingA ? ((tileK / 64) * 2 * tg::ceilDiv(tileM, 64)) * numStages : 0;
@@ -435,7 +462,7 @@ class KernelTraits
 
             // Sf B
             {
-                bool const useBlockScalingB = tg::dtypeIsBlockFmt(dtypeB);
+                bool const useBlockScalingB = tg::dtypeIsBlockFmt(dtypeMmaB);
                 // Number of columns for scaling factors of B.
                 auto const numTmemColsSfB
                     = useBlockScalingB ? ((tileK / 64) * 2 * tg::ceilDiv(tileN, 64)) * numStages : 0;
@@ -541,13 +568,20 @@ inline int32_t getSmemOffsetPerTokenSf(KernelTraits traits)
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////
 
-inline int32_t getSmemOffsetBlockAmax(KernelTraits traits)
+inline int32_t getSmemOffsetBias(KernelTraits traits)
 {
     return traits.mSmemAllocatorHelper.getChunkOffset(8);
 }
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////
 
+inline int32_t getSmemOffsetBlockAmax(KernelTraits traits)
+{
+    return traits.mSmemAllocatorHelper.getChunkOffset(9);
+}
+
+////////////////////////////////////////////////////////////////////////////////////////////////////
+
 inline int32_t isSmemAbRepurposedToGmemC(KernelTraits traits, int resIdx = 0)
 {
     // Be conscious that the index (3 + resIdx) should match the index in getSmemOffsetGmemC().
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/TmaDescriptor.h b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/TmaDescriptor.h
index 0b10632f5d..9ca92f887d 100644
--- a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/TmaDescriptor.h
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/TmaDescriptor.h
@@ -41,14 +41,14 @@ namespace tg = trtllm::gen;
 #ifdef TLLM_ENABLE_CUDA
 
 inline CUtensorMap buildNdTmaDescriptor(tg::Dtype dtype, tg::MmaKind mmaKind, std::vector<uint64_t> const& shapes,
-    std::vector<uint64_t> const& strides, int32_t tileSizeMn, int32_t tileSizeK, void* gmemAddr, bool doSwizzle = true)
+    std::vector<uint64_t> const& strides, std::vector<int32_t> const& tileShapes, void* gmemAddr, bool doSwizzle = true)
 {
     // The multiplication factor of the data padding in SMEM.
     int32_t padMultiplier = 1;
     CUtensorMap desc{};
     // The data type.
     CUtensorMapDataType tmaDataFormat{CU_TENSOR_MAP_DATA_TYPE_FLOAT32};
-    if (dtype == tg::Dtype::E4m3 || dtype == tg::Dtype::MxE4m3)
+    if (dtype == tg::Dtype::E4m3 || dtype == tg::Dtype::MxE4m3 || dtype == tg::Dtype::UE8m0)
     {
         tmaDataFormat = CU_TENSOR_MAP_DATA_TYPE_UINT8;
     }
@@ -71,15 +71,11 @@ inline CUtensorMap buildNdTmaDescriptor(tg::Dtype dtype, tg::MmaKind mmaKind, st
             padMultiplier = 2;
             tmaDataFormat = CU_TENSOR_MAP_DATA_TYPE_16U4_ALIGN16B;
         }
-        else if (mmaKind == tg::MmaKind::MxFp4NvFp4 || mmaKind == tg::MmaKind::Auto)
-        {
-            tmaDataFormat = CU_TENSOR_MAP_DATA_TYPE_16U4_ALIGN8B;
-        }
         else
         {
-            std::cerr << "Invalid dtype / mmaKind combination " << tg::dtypeToString(dtype) << "/"
-                      << tg::mmaKindToString(mmaKind) << std::endl;
-            assert(false);
+            // Note: this is used with the MMA kind MxFp4NvFp4 and also when casting to a higher-precision
+            // type such as Bfloat16 before the MMA.
+            tmaDataFormat = CU_TENSOR_MAP_DATA_TYPE_16U4_ALIGN8B;
         }
     }
     else if (dtype == tg::Dtype::Fp32)
@@ -94,24 +90,29 @@ inline CUtensorMap buildNdTmaDescriptor(tg::Dtype dtype, tg::MmaKind mmaKind, st
 
     // The swizzle type.
     CUtensorMapSwizzle swizzleType{CU_TENSOR_MAP_SWIZZLE_NONE};
-    int32_t tileKSizeInBytes = (tileSizeK * tg::dtypeGetNumBits(dtype) * padMultiplier) / /* bits */ 8;
+    int32_t fastestDimTileSizeBytes = (tileShapes[0] * tg::dtypeGetNumBits(dtype) * padMultiplier) / /* bits */ 8;
     if (doSwizzle)
     {
-        if ((tileKSizeInBytes % 128) == 0)
+        if ((fastestDimTileSizeBytes % 128) == 0)
         {
             swizzleType = CU_TENSOR_MAP_SWIZZLE_128B;
         }
-        else if ((tileKSizeInBytes % 64) == 0)
+        else if ((fastestDimTileSizeBytes % 64) == 0)
         {
             swizzleType = CU_TENSOR_MAP_SWIZZLE_64B;
         }
-        else if ((tileKSizeInBytes % 32) == 0)
+        else if ((fastestDimTileSizeBytes % 32) == 0)
         {
             swizzleType = CU_TENSOR_MAP_SWIZZLE_32B;
         }
+        else if ((fastestDimTileSizeBytes % 16) == 0 && (dtype == tg::Dtype::UE8m0 || dtype == tg::Dtype::E4m3))
+        {
+            swizzleType = CU_TENSOR_MAP_SWIZZLE_NONE;
+        }
         else
         {
-            std::cerr << "buildNdTmaDescriptor: unexpected tileKSizeInBytes " << tileKSizeInBytes << std::endl;
+            std::cerr << "buildNdTmaDescriptor: unexpected fastestDimTileSizeBytes " << fastestDimTileSizeBytes
+                      << std::endl;
             assert(false);
         }
     }
@@ -121,8 +122,9 @@ inline CUtensorMap buildNdTmaDescriptor(tg::Dtype dtype, tg::MmaKind mmaKind, st
 
     // Check shape must be in range [1, 2^32]
     int32_t dim = shapes.size();
-    // Expect 2 dimensions.
-    assert(dim == 2 || dim == 3);
+    // Expect 2 dimensions for regular gemm, 3 dimensions for batched gemm or blocked layout, and 4
+    // dimensions for batched gemm with blocked layout.
+    assert(dim == 2 || dim == 3 || dim == 4);
     // Check shape range.
     for (int32_t ii = 0; ii < dim; ++ii)
     {
@@ -147,59 +149,72 @@ inline CUtensorMap buildNdTmaDescriptor(tg::Dtype dtype, tg::MmaKind mmaKind, st
     // The number of elements in 128B.
     auto const numEltsIn128B = numEltsPerUInt32 /*4B*/ * 32;
     // The number of tile K hidden size (per token) in each block of shared memory.
-    auto const numEltsInClampedTileKSize = std::min(numEltsIn128B, tileSizeK);
+    auto const numEltsInClampedFastestTileSize = std::min(numEltsIn128B, tileShapes[0]);
 
-    // Build tile shapes.
-    std::vector<uint32_t> tileShapes(dim, 1);
-    tileShapes[0] = numEltsInClampedTileKSize; // tileSizeK
-    tileShapes[1] = tileSizeMn;                // tileSizeMn
+    // Build box dim array. If tileShapes is smaller than dim, just fill with 1s.
+    assert(static_cast<int32_t>(tileShapes.size()) <= dim);
+    std::vector<uint32_t> boxDim(dim, 1);
+    boxDim[0] = numEltsInClampedFastestTileSize;
+    for (size_t ii = 1; ii < tileShapes.size(); ++ii)
+    {
+        if (tileShapes[ii] > 256)
+        {
+            std::cerr << "buildNdTmaDescriptor: boxDim too large " << tileShapes[ii] << std::endl;
+            assert(false);
+        }
+        else
+        {
+            boxDim[ii] = tileShapes[ii];
+        }
+    }
 
     // Set tile strides to 1;
     std::vector<uint32_t> tileStrides(dim, 1);
 
     // Build the descriptor.
     CUresult result = cuTensorMapEncodeTiled(&desc, tmaDataFormat,
-        /*tensorRank=*/dim, gmemAddr, shapes.data(), stridesInBytes.data(), tileShapes.data(), tileStrides.data(),
+        /*tensorRank=*/dim, gmemAddr, shapes.data(), stridesInBytes.data(), boxDim.data(), tileStrides.data(),
         /*interleave=*/CU_TENSOR_MAP_INTERLEAVE_NONE, swizzleType,
         /*l2Promotion=*/CU_TENSOR_MAP_L2_PROMOTION_L2_128B,
         /*oobFill=*/CU_TENSOR_MAP_FLOAT_OOB_FILL_NONE);
 
     if (result != CUDA_SUCCESS)
     {
-        std::cerr << "Error: Failed to initialize the TMA descriptor " << result << std::endl;
+        std::stringstream ss;
+        ss << "Error: Failed to initialize the TMA descriptor " << result << std::endl;
 
-        std::cerr << "tmaFormat: " << static_cast<int>(tmaDataFormat) << " dim: " << dim << " gmem: " << gmemAddr
-                  << std::endl;
+        ss << "tmaFormat: " << static_cast<int>(tmaDataFormat) << " dim: " << dim << " gmem: " << gmemAddr << std::endl;
 
-        std::cerr << "Shape: ";
+        ss << "Shape: ";
         for (int ii = 0; ii < dim; ++ii)
         {
-            std::cerr << shapes[ii] << " ";
+            ss << shapes[ii] << " ";
         }
-        std::cerr << std::endl;
+        ss << std::endl;
 
-        std::cerr << "Stride: ";
+        ss << "Stride: ";
         for (int ii = 0; ii < dim - 1; ++ii)
         {
-            std::cerr << stridesInBytes[ii] << " ";
+            ss << stridesInBytes[ii] << " ";
         }
-        std::cerr << std::endl;
+        ss << std::endl;
 
-        std::cerr << "tileShapes: ";
+        ss << "tileShapes: ";
         for (int ii = 0; ii < dim; ++ii)
         {
-            std::cerr << tileShapes[ii] << " ";
+            ss << boxDim[ii] << " ";
         }
-        std::cerr << std::endl;
+        ss << std::endl;
 
-        std::cerr << "tileStrides: ";
+        ss << "tileStrides: ";
         for (int ii = 0; ii < dim; ++ii)
         {
-            std::cerr << tileStrides[ii] << " ";
+            ss << tileStrides[ii] << " ";
         }
-        std::cerr << std::endl;
-        std::cerr << "swizzleType: " << int(swizzleType) << std::endl;
-        assert(false);
+        ss << std::endl;
+        ss << "swizzleType: " << int(swizzleType) << std::endl;
+        ss << "(in " << __FILE__ << ":" << __LINE__ << ")" << std::endl;
+        throw std::runtime_error(ss.str());
     }
 
     return desc;
@@ -267,41 +282,42 @@ inline CUtensorMap buildSfTmaDescriptor(tg::Dtype dtype, std::vector<uint64_t> c
 
     if (result != CUDA_SUCCESS)
     {
-        std::cerr << "Error: Failed to initialize the TMA descriptor for SF " << result << std::endl;
+        std::stringstream ss;
+        ss << "Error: Failed to initialize the TMA descriptor for SF " << result << std::endl;
 
-        std::cerr << "tmaFormat: " << static_cast<int>(tmaDataFormat) << " dim: " << dim << " gmem: " << gmemAddr
-                  << std::endl;
+        ss << "tmaFormat: " << static_cast<int>(tmaDataFormat) << " dim: " << dim << " gmem: " << gmemAddr << std::endl;
 
-        std::cerr << "shape:";
+        ss << "shape:";
         for (uint32_t shape_i : shapes)
         {
-            std::cerr << " " << shape_i;
+            ss << " " << shape_i;
         }
-        std::cerr << std::endl;
+        ss << std::endl;
 
-        std::cerr << "stridesInBytes:";
+        ss << "stridesInBytes:";
         for (uint32_t stride_i : stridesInBytes)
         {
-            std::cerr << " " << stride_i;
+            ss << " " << stride_i;
         }
-        std::cerr << std::endl;
+        ss << std::endl;
 
-        std::cerr << "tileShapes:";
+        ss << "tileShapes:";
         for (uint32_t tileShape_i : tileShapes)
         {
-            std::cerr << " " << tileShape_i;
+            ss << " " << tileShape_i;
         }
-        std::cerr << std::endl;
+        ss << std::endl;
 
-        std::cerr << "tileStrides:";
+        ss << "tileStrides:";
         for (uint32_t tileStride_i : tileStrides)
         {
-            std::cerr << " " << tileStride_i;
+            ss << " " << tileStride_i;
         }
-        std::cerr << std::endl;
+        ss << std::endl;
 
-        std::cerr << "swizzleType: " << int(swizzleType) << std::endl;
-        assert(false);
+        ss << "swizzleType: " << int(swizzleType) << std::endl;
+        ss << "(in " << __FILE__ << ":" << __LINE__ << ")" << std::endl;
+        throw std::runtime_error(ss.str());
     }
 
     return desc;
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/config.json b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/config.json
index cf40d3e452..8f553bab03 100644
--- a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/config.json
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/config.json
@@ -180,6 +180,7 @@
             "dtypeC": ["bf16", "fp16", "e2m1"],
             "listN": "8,8",
             "numExperts": 2,
+            "numStagesMma": 1,
             "tileScheduler": "static"
         },
         {
@@ -191,6 +192,7 @@
             "dtypeC": ["bf16", "fp16", "e4m3"],
             "listN": "8,8",
             "numExperts": 2,
+            "numStagesMma": 1,
             "tileScheduler": "static"
         },
         {
@@ -202,6 +204,7 @@
             "dtypeC": ["bf16", "fp16", "e4m3"],
             "listN": "8,8",
             "numExperts": 2,
+            "numStagesMma": 2,
             "tileScheduler": "static"
         },
         {
@@ -209,15 +212,17 @@
             "_template": "BatchedGemmFp4LowLatency",
             "routeAct": true,
             "fusedAct": true,
+            "sfLayoutB": "linear",
             "useRoutingScalesOnInput": false,
             "useUnrollLoop2xForMma": [true, false],
             "dtypeC": "e2m1",
             "numTokens": 2,
             "numExperts": 2,
+            "numStagesMma": 1,
             "tileScheduler": "static"
         },
         {
-            "_comment": "DS_FP4_FC2",
+            "_comment": "DS_FP4_FC2_persistSched",
             "_template": "BatchedGemmFp4LowLatency",
             "routeAct": false,
             "fusedAct": false,
@@ -227,10 +232,23 @@
             "numTokens": 2,
             "numExperts": 2,
             "numStagesMma": 2,
-            "tileScheduler": ["persistent", "static"]
+            "tileScheduler": "persistent"
+        },
+        {
+            "_comment": "DS_FP4_FC2_staticSched",
+            "_template": "BatchedGemmFp4LowLatency",
+            "routeAct": false,
+            "fusedAct": false,
+            "useRoutingScalesOnInput": false,
+            "useUnrollLoop2xForMma": [true, false],
+            "dtypeC": "bf16",
+            "numTokens": 2,
+            "numExperts": 2,
+            "numStagesMma": 1,
+            "tileScheduler": "static"
         },
         {
-            "_comment": "DS_FP4_TP8_EP1_FC2",
+            "_comment": "DS_FP4_TP8_EP1_FC2_persistSched",
             "_template": "BatchedGemmFp4LowLatency",
             "routeAct": false,
             "fusedAct": false,
@@ -242,7 +260,22 @@
             "numStagesMma": 2,
             "tileK": 256,
             "numStages": 6,
-            "tileScheduler": ["persistent", "static"]
+            "tileScheduler": "persistent"
+        },
+        {
+            "_comment": "DS_FP4_TP8_EP1_FC2_staticSched",
+            "_template": "BatchedGemmFp4LowLatency",
+            "routeAct": false,
+            "fusedAct": false,
+            "useRoutingScalesOnInput": false,
+            "useUnrollLoop2xForMma": [true, false],
+            "dtypeC": "bf16",
+            "numTokens": 2,
+            "numExperts": 2,
+            "numStagesMma": 1,
+            "tileK": 256,
+            "numStages": 6,
+            "tileScheduler": "static"
         },
         {
             "_comment": "DS_FP8_FC1",
@@ -259,7 +292,7 @@
             "tileScheduler": "static"
         },
         {
-            "_comment": "DS_FP8_FC2",
+            "_comment": "DS_FP8_FC2_persistent",
             "_template": "BatchedGemmDeepSeekFp8LowLatency",
             "routeAct": false,
             "fusedAct": false,
@@ -270,7 +303,21 @@
             "numExperts": 2,
             "numStages": 8,
             "numStagesMma": 4,
-            "tileScheduler": ["persistent", "static"]
+            "tileScheduler": "persistent"
+        },
+        {
+            "_comment": "DS_FP8_FC2_static",
+            "_template": "BatchedGemmDeepSeekFp8LowLatency",
+            "routeAct": false,
+            "fusedAct": false,
+            "useRoutingScalesOnInput": false,
+            "useUnrollLoop2xForMma": [true, false],
+            "dtypeC": "bf16",
+            "numTokens": 2,
+            "numExperts": 2,
+            "numStages": 8,
+            "numStagesMma": 2,
+            "tileScheduler": "static"
         },
         {
             "_comment": "DS_FP8_FC1_16_THROUGHPUT",
@@ -407,7 +454,7 @@
             "tileScheduler": "static"
         },
         {
-            "_comment": "Llama4_FP8_FC2",
+            "_comment": "Llama4_FP8_FC2_persistent",
             "_template": "BatchedGemmPerTensorScalingFp8LowLatency",
             "routeAct": false,
             "fusedAct": false,
@@ -418,22 +465,38 @@
             "numExperts": 2,
             "numStages": 3,
             "numStagesMma": 2,
-            "tileScheduler": ["persistent", "static"]
+            "tileScheduler": "persistent"
+        },
+        {
+            "_comment": "Llama4_FP8_FC2_static",
+            "_template": "BatchedGemmPerTensorScalingFp8LowLatency",
+            "routeAct": false,
+            "fusedAct": false,
+            "useRoutingScalesOnInput": false,
+            "useUnrollLoop2xForMma": [true, false],
+            "dtypeC": "bf16",
+            "numTokens": 2,
+            "numExperts": 2,
+            "numStages": 3,
+            "numStagesMma": 1,
+            "tileScheduler": "static"
         },
         {
             "_comment": "MxE2m1_FC1",
             "_template": "BatchedGemmMxE2m1MxE4m3LowLatency",
             "routeAct": true,
             "fusedAct": true,
+            "sfLayoutB": "linear",
             "useRoutingScalesOnInput": false,
             "useUnrollLoop2xForMma": [true, false],
             "dtypeC": "mxe4m3",
             "numTokens": 2,
             "numExperts": 2,
+            "numStagesMma": 1,
             "tileScheduler": "static"
         },
         {
-            "_comment": "MxE2m1_FC2",
+            "_comment": "MxE2m1_FC2_persistent",
             "_template": "BatchedGemmMxE2m1MxE4m3LowLatency",
             "routeAct": false,
             "fusedAct": false,
@@ -443,10 +506,23 @@
             "numTokens": 2,
             "numExperts": 2,
             "numStagesMma": 2,
-            "tileScheduler": ["persistent", "static"]
+            "tileScheduler": "persistent"
+        },
+        {
+            "_comment": "MxE2m1_FC2_static",
+            "_template": "BatchedGemmMxE2m1MxE4m3LowLatency",
+            "routeAct": false,
+            "fusedAct": false,
+            "useRoutingScalesOnInput": false,
+            "useUnrollLoop2xForMma": [true, false],
+            "dtypeC": "bf16",
+            "numTokens": 2,
+            "numExperts": 2,
+            "numStagesMma": 1,
+            "tileScheduler": "static"
         },
         {
-            "_comment": "MxE2m1_TP8_EP1_FC2",
+            "_comment": "MxE2m1_TP8_EP1_FC2_persistent",
             "_template": "BatchedGemmMxE2m1MxE4m3LowLatency",
             "routeAct": false,
             "fusedAct": false,
@@ -458,7 +534,122 @@
             "numStagesMma": 2,
             "tileK": 256,
             "numStages": 4,
-            "tileScheduler": ["persistent", "static"]
+            "tileScheduler": "persistent"
+        },
+        {
+            "_comment": "MxE2m1_TP8_EP1_FC2_static",
+            "_template": "BatchedGemmMxE2m1MxE4m3LowLatency",
+            "routeAct": false,
+            "fusedAct": false,
+            "useRoutingScalesOnInput": false,
+            "useUnrollLoop2xForMma": [true, false],
+            "dtypeC": "bf16",
+            "numTokens": 2,
+            "numExperts": 2,
+            "numStagesMma": 1,
+            "tileK": 256,
+            "numStages": 4,
+            "tileScheduler": "static"
+        },
+        {
+            "_comment": "Qwen3_TP1_EP8_FC1",
+            "_template": "BatchedGemmFp4LowLatency",
+            "routeAct": true,
+            "fusedAct": true,
+            "sfLayoutB": "linear",
+            "dtypeC": "e2m1",
+            "numStages": 4,
+            "numStagesMma": 1,
+            "tileK": 512,
+            "tileScheduler": "static",
+            "useUnrollLoop2xForMma": [true]
+        },
+        {
+            "_comment": "Qwen3_TP1_EP8_FC2",
+            "_template": "BatchedGemmFp4LowLatency",
+            "routeAct": false,
+            "fusedAct": false,
+            "dtypeC": "bf16",
+            "numStages": 4,
+            "numStagesMma": 1,
+            "tileK": 512,
+            "tileScheduler": "static",
+            "useUnrollLoop2xForMma": [false]
+        },
+        {
+            "_comment": "Qwen3_TP2_EP4_FC1",
+            "_template": "BatchedGemmFp4LowLatency",
+            "routeAct": true,
+            "fusedAct": true,
+            "sfLayoutB": "linear",
+            "dtypeC": "e2m1",
+            "numStages": 4,
+            "numStagesMma": 1,
+            "tileK": 512,
+            "tileScheduler": "static",
+            "useUnrollLoop2xForMma": [true]
+        },
+        {
+            "_comment": "Qwen3_TP2_EP4_FC2",
+            "_template": "BatchedGemmFp4LowLatency",
+            "routeAct": false,
+            "fusedAct": false,
+            "dtypeC": "bf16",
+            "numStages": 4,
+            "numStagesMma": 2,
+            "tileK": 512,
+            "tileScheduler": "persistent",
+            "useUnrollLoop2xForMma": [false]
+        },
+        {
+            "_comment": "Qwen3_TP4_EP2_FC1",
+            "_template": "BatchedGemmFp4LowLatency",
+            "routeAct": true,
+            "fusedAct": true,
+            "sfLayoutB": "linear",
+            "dtypeC": "e2m1",
+            "numStages": 4,
+            "numStagesMma": 1,
+            "tileK": 512,
+            "tileScheduler": "static",
+            "useUnrollLoop2xForMma": [true]
+        },
+        {
+            "_comment": "Qwen3_TP4_EP2_FC2",
+            "_template": "BatchedGemmFp4LowLatency",
+            "routeAct": false,
+            "fusedAct": false,
+            "dtypeC": "bf16",
+            "numStages": 4,
+            "numStagesMma": 2,
+            "tileK": 512,
+            "tileScheduler": "persistent",
+            "useUnrollLoop2xForMma": [false]
+        },
+        {
+            "_comment": "Qwen3_TP8_EP1_FC1",
+            "_template": "BatchedGemmFp4LowLatency",
+            "routeAct": true,
+            "fusedAct": true,
+            "sfLayoutB": "linear",
+            "dtypeC": "e2m1",
+            "numStages": 4,
+            "numStagesMma": 1,
+            "tileK": 512,
+            "tileScheduler": "static",
+            "useUnrollLoop2xForMma": [true]
+        },
+        {
+            "_comment": "Qwen3_TP8_EP1_FC2",
+            "_template": "BatchedGemmFp4LowLatency",
+            "routeAct": false,
+            "fusedAct": false,
+            "dtypeC": "bf16",
+            "numStages": 4,
+            "numStagesMma": 2,
+            "tileK": 256,
+            "tileScheduler": "persistent",
+            "useUnrollLoop2xForMma": [false]
         }
     ]
 }
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..f80b9172e6
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:025a21a07a797de485c2c86bfc5dbbe1fb5c494466346a207e7fc8ac266f60b8
+size 583355
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..de97381a21
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5d0f7bcd89357e66b7c75ec67ccb481de87d8c73898ee8f81672b0eb35830769
+size 691616
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..ed326410aa
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c73f79940b8691ecc3a14ce254bd64afced1824a79ac96f2e305dedfebe8f7b5
+size 606299
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..5064016786
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x256u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c7d783dc7ece8cea35d9bf83a7c8eea3986d9f29c3cd67389433d2357ade2d74
+size 729162
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..20888a1d49
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a4a8773b5462296ac6b1fb65c34e5fec54be4dd41e3b52c50e463c436f616d12
+size 552325
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp
new file mode 100644
index 0000000000..a609d1ffe2
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:afb26c01c09a14d5fd5243d018164e3cb47f6acb26b380810bd24ea20b4deef3
+size 549397
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..6c75e50a77
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:19f7b7a44abef9ab58ead44ea82570c2669c41b2bf9dc97300a1cdc732aff6dd
+size 688754
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..0811ac534f
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:532b00782f06b21405eae6c207cb539494b552d7db05bcecd96bf872d4a586fd
+size 578475
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp
new file mode 100644
index 0000000000..024e1e7caf
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fd8cf63739a7968321556b7980ce56577f9692921a7f62043295be7e0f9f19eb
+size 575499
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..dcb2ed07e4
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3cc777b009a1fe7fcf4f5c2b794c0eaca2943e8f190c94fc4d3bb30c49348f85
+size 722256
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x128x128_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x128x128_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..43a9dab09d
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x128x128_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f072615f04e0d0678e70294c0d6b7486e225d2382bad8f8427390a125bb48236
+size 610163
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x128x128_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x128x128_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..27af07cab1
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x128x128_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:83bb1c43a309a3c0ce1f56cc239e517a7596a4cbdd59c16adec1ca9fb708043c
+size 779992
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x128x128u2_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x128x128u2_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..4cf4506310
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x128x128u2_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8994ba4c2ca7dd0bf592da450e1404441f358657f7f60c4cdd56b56f9c448131
+size 628668
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x128x128u2_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x128x128u2_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..da87049c7f
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x128x128u2_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5cdc0f6feff2b0ea40dfb73159dc2a5038a2f46b328d057da200bd76d1fec044
+size 806784
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x16x128_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x16x128_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..68b132b2b6
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x16x128_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:81396786711ebd423e67d113dc1f1d2d963385b1f56b56d51448070b538678d2
+size 501179
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x16x128_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x16x128_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..6d2239f8a0
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x16x128_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0d82db55cdd10e897a4e3f7778c09c10bf029a69a86a80da92baebe7a6dfcaac
+size 632034
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x16x128u2_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x16x128u2_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..f632f63e44
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x16x128u2_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c23c02a534806734e5522c54cf9f6ddc112df09c35a0834439796f357ea2c700
+size 518895
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x16x128u2_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x16x128u2_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..4485f5c7cc
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x16x128u2_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6dca1a3d18d04e8d20b41bea46729faaf36d22e1aec5e75b2dc9d8c074dfca5b
+size 661244
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x32x128_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x32x128_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..104fa0c8c1
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x32x128_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f66cc836ac67fbf37e27c790d93fd56ace7f84e45e1e8e709bc92449f171c40b
+size 518249
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x32x128_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x32x128_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..54a234683b
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x32x128_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:44a920497d1cac7d3c2e27898758f4a069b2e40b16851e1c671a57580a6de613
+size 653742
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x32x128u2_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x32x128u2_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..27487413df
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x32x128u2_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5de768877c4368ea3798f6410d5d66cd5010e7debc14fee453941f823cca1d75
+size 535963
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x32x128u2_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x32x128u2_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..41805ff56f
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x32x128u2_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aed58a7fca6462cd5f1f15181c4818071dd99fe7b436b570132c6cc3df6e97ed
+size 682112
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x64x128_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x64x128_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..36cfaeacd4
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x64x128_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5f7294cdde3778c06141ce0ae68e293ec6b9ca990373e452c01f91852cbbb556
+size 554015
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x64x128_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x64x128_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..81de55d568
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x64x128_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8cbbf9129f822e95d3fae46330460eb5b50878e12c5f3d495102edb04fa7ca69
+size 699918
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x64x128u2_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x64x128u2_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..8da3ce0b74
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x64x128u2_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:28086a74790bb3e11722786569b95be426bf3ada28a82a36730dd3465d904fa7
+size 571731
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x64x128u2_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x64x128u2_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..a63d7844f7
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x64x128u2_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:92fa64e4f820b2f5d1b4db1864220d36231c2a6451c8a93b5aa13a3cd299caf4
+size 729028
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..94c748b3ce
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a8f2365a13656aaf10d935265da62b26ff30019fd151296a2c1dea1ee633d1fe
+size 539851
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin.cpp
new file mode 100644
index 0000000000..5db3fd96d6
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9d42abe6ddcac40aaa119421f50038adb4edf9f4e12931ba8528a858bd4de0ca
+size 514477
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..7859a3975e
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8b3e59916d41f8e66fa2d474819a031d2a24caaba31485dce0a62d5e12e2d24a
+size 629660
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..8fc8d104fb
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:917ef5d93cf7cfbd76f3d32b3c051f4e119631a0324801a4b8684bf98157857c
+size 559193
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin.cpp
new file mode 100644
index 0000000000..d5d56c56a8
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:330cad5e70b49cf039b5931e9fa6686298bd04d9416e7684aa5f31c4e5e46444
+size 533821
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..1ac491c269
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_persistent_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:82707942c797cba2d1352e94fa594aca0a8b1c3c07d99646c41227a83f740766
+size 651864
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..7f8299ac92
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f7f0619d6293f8448940ba954221fe2c841541bd8199fc943983eb6acd41ed61
+size 420605
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp
new file mode 100644
index 0000000000..3bbf5ee43b
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7882097269f622d56d99f704b42a0331d1fdc59442e803ff9c8f836b67c7d7d2
+size 417333
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..8640661908
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:34e40aba2b816a6bee4e8f30bc758d7ed3438b290341d6834c7985232f3c4760
+size 504443
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..87b13fa745
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fbf07013056fcb9c58a4a90fa2783eb38a3ee311e2ec4d2a5fb7aa855399a389
+size 443893
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp
new file mode 100644
index 0000000000..0bb069d3f8
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:15e4735cc2d2e2bf3fdb10698e20303a532939acdfcb6c8c691b2a2d569663a6
+size 440671
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..b2669c3acf
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2a9d9b821d0dcb825bed9911cdd556ed063ba48b398294eb4848159a2d11ae34
+size 526055
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..21889830f0
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8faa5ad1e00d5e9bfd17f288934536ad440d4655c5efac5e6a1fe01e7be245fa
+size 563777
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..aecb4e1dd9
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x256_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bb0133260a01b83ed44b6a0b150dbcad2857f575a70476c41da2cca6a9b35f55
+size 736122
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x256u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x256u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..6b74f349ae
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x256u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1acd2b5b82fd69f3cf568ff96ace34ac5ed120785e5c5a5b8f45dccb7a589422
+size 591605
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x256u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x256u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..25066e0555
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x256u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1df67c891567c1b074da7b6dae1b3b819bde680cd35b2cf5ab89a750b8cc943a
+size 761582
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..9a82468a76
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3e3bc3b8b35eba8111d94ab7d195b8b731cc9b5f441bda0bf7784e0fdec1cc09
+size 610349
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..539d9762d0
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7770d7a52f7ec065e1f3df861c273edf6f7e1cc677c36ac11c0352b50be83649
+size 739970
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..2e3d6fc33d
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:86d84a8dbf978b75d87959e2cbcc50f1863d8f2c771bf26c5c4ad47996f81e18
+size 640150
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..226404950d
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Bfloat16_MxE2m1MxE4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_persistent_batchN_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:385fa25c09e6dc066957de238fea4f0636ee52cc90f7ad456c61f68390b13aad
+size 766516
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..eda32eb064
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ef6738ce1712ecebe6e717e09bd7d78b324f828a4b3f4df2898511275383aad4
+size 600993
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp
new file mode 100644
index 0000000000..108550c5ee
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f4e0e4a92d078fc240e9006ae271cc08a91f214d31cc40269affe86f83b664ec
+size 555211
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..a245c9b0e8
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7e5b78e9b3db3a45609fba97e9df3c960f5f380e6a59bfddb2423f48908bf51f
+size 626404
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp
new file mode 100644
index 0000000000..7820f4b9a7
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E2m1_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b6d33c4ccff36bae9207ff93168d3607d5b0681bc586d50b8c6995b409661403
+size 581313
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x128x128_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x128x128_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..7e07a1dcc7
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x128x128_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d730384eeb2bd8f97d3f0daeabce52e0ea4ed80cac954d9755f51108b0ffce97
+size 841702
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x128x128u2_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x128x128u2_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..3bae4036eb
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x128x128u2_epilogueTile64x128_mma64x128x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:20642ded83c712aeeb9ba418353f6fa7552c4622777c61d128253e0e598b85e1
+size 859416
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x16x128_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x16x128_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..0114b7a888
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x16x128_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c8805e951d49c58123db07253b867803d3ae30942269c80915e7ea81eeffba66
+size 524185
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x16x128u2_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x16x128u2_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..093b53806e
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x16x128u2_epilogueTile64x16_mma64x16x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:15c42189f82f2ea343576b589a1c28748d0ab3af0d04a11d5925f1cd97c4e23e
+size 542689
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x32x128_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x32x128_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..689d1fcf71
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x32x128_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8f5a33ba3b1201b9884329362f5192ab4d394d26b596b5d96f9ab64535f9426b
+size 565625
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x32x128u2_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x32x128u2_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..81365ccdbb
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x32x128u2_epilogueTile64x32_mma64x32x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ecbdc8d75681767b45d8f967a7f13dc9c852fa131c274b6c22dac829a0b4a919
+size 582551
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x64x128_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x64x128_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..2097d2c376
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x64x128_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b496add3396b71793dbe998f4b1579c1bcae810286df86b0351f12c4197a4697
+size 633262
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x64x128u2_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x64x128u2_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..bf7f917c8d
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x64x128u2_epilogueTile64x64_mma64x64x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:246d7c59b39586b37dd675292254fe90654e5d158ef24aef8ce04d4543c55aed
+size 650976
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..6a1f9358b2
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:868f0fedc4e99546431d08068f43cdd9c3c2f523ed922844b66f4ce83717c402
+size 558219
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin.cpp
new file mode 100644
index 0000000000..5f523f2729
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d0463544c6a80fd23dbdc767dfc96f512b2757dec2d976c917a9850a16f9e6f5
+size 527297
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..4994666099
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_routeLdgsts_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3fd00a01e561343de9f8bd212b5e6da151e55e68cb12e37aa2ee0edb7cf3d76d
+size 575243
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin.cpp
new file mode 100644
index 0000000000..2828d8fb73
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2cb24d3bc83aabe4c4ae6bac627a70fbc5cfc537f917a6c93b8d6830e1e0c373
+size 545849
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..07dd143888
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9c805a4f096f9f792b6402c607565f2e88fc1a56344e654618c3bbe1b615e01c
+size 392855
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp
new file mode 100644
index 0000000000..fc76a27474
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:08067655e5ad721255bb86ec17617c7217ef9dc1165d1ce4bcc11c5a86dec681
+size 415845
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..908071b19a
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c4cc962ca87e8e42ccf3aab87d2c921cf747ff1cc8d2cb8e9b339491a47122c0
+size 418563
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp
new file mode 100644
index 0000000000..c453b40432
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_E4m3_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:342721dcf5114816a32a15d31e24711a2aadd52ea4eb578695c502324786f163
+size 439973
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Fp16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Fp16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp
new file mode 100644
index 0000000000..0b5e17dbcf
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Fp16_E2m1E2m1_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f11f3bc7b0c137577c335e4762068a0de4785a544d85f2575b787360c300f0a8
+size 548601
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Fp16_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Fp16_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp
new file mode 100644
index 0000000000..af47072c14
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Fp16_E2m1E2m1_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x64_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2a35b8ab05817e3168b36002a8c0bfa45ff6cbe09d3a2db4af8be4dfb95fb6e6
+size 574701
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Fp16_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Fp16_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin.cpp
new file mode 100644
index 0000000000..4ba5eb66d1
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Fp16_E4m3E4m3_Fp32_tile128x8x128_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2c4f2e7ba6b10a99f386253c6b9315620804590f2feae7dd78351c3fce34d9ec
+size 513681
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Fp16_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Fp16_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin.cpp
new file mode 100644
index 0000000000..a5ae523b14
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Fp16_E4m3E4m3_Fp32_tile128x8x128u2_epilogueTile64x8_mma64x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_dsFp8_batchN_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:86179f2a7d2ddbffcb8b40a49f5babc31dc6dd80acff882a67a1eae40d396b24
+size 532233
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Fp16_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Fp16_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp
new file mode 100644
index 0000000000..b9c02816a7
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Fp16_E4m3E4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:59566acf5116dd2221842073c5fcea6bcf70eb5ee29b14482e5d4efd33ebadb4
+size 415745
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Fp16_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Fp16_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp
new file mode 100644
index 0000000000..1cdf0ce039
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_Fp16_E4m3E4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:41933fab6978bc725dad91ccd0539a25f804f1e579a0360d3d6289eab0e076de
+size 439873
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_MxE4m3_MxE2m1MxE4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_MxE4m3_MxE2m1MxE4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..f6e7eff70c
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_MxE4m3_MxE2m1MxE4m3_Fp32_tile128x8x512_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bf376b0111b28c57f20cdd44a5edaa98edd364da22b194dcfa4188e8247068c7
+size 625426
diff --git a/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_MxE4m3_MxE2m1MxE4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin.cpp b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_MxE4m3_MxE2m1MxE4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin.cpp
new file mode 100644
index 0000000000..d0ad1d6f84
--- /dev/null
+++ b/cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/cubins/BatchedGemmKernel_MxE4m3_MxE2m1MxE4m3_Fp32_tile128x8x512u2_epilogueTile128x8_mma128x8x32_cluster1x1x1_16dp256bit_TN_transposeMmaOutput_batchN_routeLdgsts_silu_dynamic_sm100a_cubin.cpp
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fef76304ee09e3ca32b459db98ef67f3a5cd93e8526c208115227b97eacaf22e
+size 659520
diff --git a/tests/unittest/_torch/thop/test_moe.py b/tests/unittest/_torch/thop/test_moe.py
index db76f62af1..dcae98cac5 100644
--- a/tests/unittest/_torch/thop/test_moe.py
+++ b/tests/unittest/_torch/thop/test_moe.py
@@ -593,8 +593,6 @@ def test_moe_fp8(num_tokens, expert_info, hidden_size, intermediate_size):
     assert num_experts % n_groups == 0
     assert num_experts % 4 == 0
     assert top_k < (top_k_groups * num_experts / n_groups)
-    assert hidden_size % 128 == 0
-    assert intermediate_size % 128 == 0
 
     expert_logits = torch.randn((num_tokens, num_experts),
                                 device='cuda').to(torch.float)
@@ -668,9 +666,9 @@ def check_accuracy(a, b, atol, rtol, percent):
     reason="The kernel only supports Blackwell. Current SM is %d." %
     getSMVersion(),
 )
-@pytest.mark.parametrize("num_tokens", [1, 2, 16, 64, 1024, 4096])
+@pytest.mark.parametrize("num_tokens", [1, 1024, 4096])
 @pytest.mark.parametrize("hidden_size", [1024])
-@pytest.mark.parametrize("intermediate_size", [1024])
+@pytest.mark.parametrize("intermediate_size", [1024, 768, 384, 192])
 @pytest.mark.parametrize(
     "routing_info",
     [
@@ -743,8 +741,6 @@ def test_moe_fp4(num_tokens, hidden_size, intermediate_size, routing_info):
 
     assert top_k <= num_experts
     assert top_k <= 8
-    assert hidden_size % 128 == 0
-    assert intermediate_size % 128 == 0
     if (top_k_groups is not None) and (n_groups is not None):
         assert top_k_groups <= 4
         assert num_experts > n_groups
@@ -995,8 +991,6 @@ def test_moe_fp8_per_tensor_scale(num_tokens, expert_info, hidden_size,
     assert n_groups == 0 or num_experts % n_groups == 0
     assert num_experts % 4 == 0
     assert n_groups == 0 or top_k < (top_k_groups * num_experts / n_groups)
-    assert hidden_size % 128 == 0
-    assert intermediate_size % 128 == 0
 
     expert_logits = torch.randn((num_tokens, num_experts),
                                 device='cuda').to(torch.float)