[Refactor] Remove Unused Env VLLM_ENABLE_MOE_ALIGN_BLOCK_SIZE_TRITON (#20334)

Signed-off-by: yewentao256 <zhyanwentao@126.com>

Author: yewentao256

vllm/envs.py

@@ -104,7 +104,6 @@
     VLLM_SERVER_DEV_MODE: bool = False
     VLLM_V1_OUTPUT_PROC_CHUNK_SIZE: int = 128
     VLLM_MLA_DISABLE: bool = False
-    VLLM_ENABLE_MOE_ALIGN_BLOCK_SIZE_TRITON: bool = False
     VLLM_RAY_PER_WORKER_GPUS: float = 1.0
     VLLM_RAY_BUNDLE_INDICES: str = ""
     VLLM_CUDART_SO_PATH: Optional[str] = None
@@ -769,12 +768,6 @@ def get_vllm_port() -> Optional[int]:
     "VLLM_MLA_DISABLE":
     lambda: bool(int(os.getenv("VLLM_MLA_DISABLE", "0"))),
 
-    # If set, vLLM will use the Triton implementation of moe_align_block_size,
-    # i.e. moe_align_block_size_triton in fused_moe.py.
-    "VLLM_ENABLE_MOE_ALIGN_BLOCK_SIZE_TRITON":
-    lambda: bool(int(os.getenv("VLLM_ENABLE_MOE_ALIGN_BLOCK_SIZE_TRITON", "0"))
-                 ),
-
     # Number of GPUs per worker in Ray, if it is set to be a fraction,
     # it allows ray to schedule multiple actors on a single GPU,
     # so that users can colocate other actors on the same GPUs as vLLM.

vllm/model_executor/layers/fused_moe/moe_align_block_size.py

@@ -94,7 +94,6 @@ def moe_align_block_size_stage4(
 
 # Triton implementation based on:
 # https://github.com/sgl-project/sglang/commit/ba5112ff691d791a9e38c6c71f59324a5fcb49d0
-# TODO(wentao): Deprecated this function in the future.
 def moe_align_block_size_triton(
     topk_ids: torch.Tensor,
     num_experts: int,