lint

Signed-off-by: czhu-cohere <conway.zhu@cohere.com>

Author: czhu-cohere

benchmarks/kernels/weight_shapes.py

@@ -96,9 +96,9 @@
         ([1408, 2048], 0),
     ],
     "CohereLabs/c4ai-command-a-03-2025": [
-        ([12288, 36864*2], 1),  # gate_up_proj
+        ([12288, 73728], 1),    # gate_up_proj
         ([36864, 12288], 0),    # down_proj
         ([12288,  14336], 1),   # qkv_proj
         ([12288, 12288], 0),    # o_proj
-    ]
+    ],
 }

csrc/quantization/machete/generate.py

@@ -508,7 +508,8 @@ def generate():
 
     sch_config_overrides = {
         #### M = 1-16
-        None: dict(
+        None:
+        dict(
             kernel_schedule=MixedInputKernelScheduleType.TmaWarpSpecialized,
             epilogue_schedule=TmaCoop,
             tile_scheduler=TileSchedulerType.Default,
@@ -518,11 +519,12 @@ def generate():
     # For now we use the same heuristic for all types
     # Heuristic is currently tuned for H100s
     default_heuristic = [
-        (cond, ScheduleConfig(
-            *tile_config,
-            **sch_config_overrides.get(cond, sch_common_params)  # type: ignore
-        ))
-        for cond, tile_config in default_tile_heuristic_config.items()
+        (
+            cond, ScheduleConfig(
+                *tile_config,
+                **sch_config_overrides.get(cond,
+                                           sch_common_params)  # type: ignore
+        )) for cond, tile_config in default_tile_heuristic_config.items()
     ]
 
     def get_unique_schedules(heuristic: dict[str, ScheduleConfig]):

csrc/quantization/machete/machete_mm_kernel.cuh

@@ -40,8 +40,7 @@ using namespace cute;
 //   we compute the transpose to move it to the left-hand side.
 template <typename ElementA_, typename ElementB_, typename ElementD_,
           typename AccumulatorT, typename GroupScaleT, typename GroupZeroT,
-          typename ChannelScaleT, typename TokenScaleT,
-          typename ScheduleConfig>
+          typename ChannelScaleT, typename TokenScaleT, typename ScheduleConfig>
 struct MacheteKernelTemplate {
   static constexpr bool with_C = false;  // not ever used
   static constexpr bool with_group_scales = !std::is_same_v<GroupScaleT, void>;