[Perf] Optimize moe_align_block_size CUDA kernel (#19572)

Signed-off-by: yewentao256 <zhyanwentao@126.com>
Co-authored-by: mgoin <mgoin64@gmail.com>
This commit is contained in:
Wentao Ye
2025-06-17 11:49:26 -07:00
committed by GitHub
co-authored by mgoin
parent ca94d7fa00
commit ffb2cd6b54
7 changed files with 386 additions and 365 deletions
-9
View File
@@ -1524,15 +1524,6 @@ def moe_align_block_size(topk_ids: torch.Tensor, num_experts: int,
num_tokens_post_pad)
def sgl_moe_align_block_size(topk_ids: torch.Tensor, num_experts: int,
block_size: int, sorted_token_ids: torch.Tensor,
experts_ids: torch.Tensor,
num_tokens_post_pad: torch.Tensor) -> None:
torch.ops._moe_C.sgl_moe_align_block_size(topk_ids, num_experts,
block_size, sorted_token_ids,
experts_ids, num_tokens_post_pad)
def moe_wna16_gemm(input: torch.Tensor, output: torch.Tensor,
b_qweight: torch.Tensor, b_scales: torch.Tensor,
b_qzeros: Optional[torch.Tensor],