A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
[Doc] Fixed shape description for fused_batched_moe.py (#25668)
Signed-off-by: Egor <e.a.krivov@gmail.com>
Egor committed
Oct 3, 2025 at 13:00 UTC
aa5053e3feb26c7de244b204bccc8d403b5b3df5
1 file changed
+1
-1
vllm/model_executor/layers/fused_moe/fused_batched_moe.py
+1
-1
@@ -355,7 +355,7 @@ def batched_triton_kernel(
355
356
def invoke_moe_batched_triton_kernel(
357
A: torch.Tensor, # [E, max_tokens, K]
358
- B: torch.Tensor, # [E, K, N]
358
+ B: torch.Tensor, # [E, N, K]
359
C: torch.Tensor, # [E, max_tokens, N]
360
expert_num_tokens: torch.Tensor, # [E]
361
compute_type: tl.dtype,