[Bugfix] Handle num_expert_group=None in flashinfer block-scale FP8 MoE (#34494)
Signed-off-by: haosdent <haosdent@gmail.com>
This commit is contained in:
@@ -201,6 +201,7 @@ def flashinfer_fused_moe_blockscale_fp8(
|
||||
) -> torch.Tensor:
|
||||
from vllm.utils.flashinfer import flashinfer_trtllm_fp8_block_scale_moe
|
||||
|
||||
num_expert_group = num_expert_group if num_expert_group is not None else 0
|
||||
topk_group = topk_group if topk_group is not None else 0
|
||||
assert top_k <= global_num_experts
|
||||
assert top_k <= 10
|
||||
|
||||
Reference in New Issue
Block a user