EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

Set num_fused_shared_experts as num_shared_experts when shared_experts fusion is not disabled (#6736)

Browse Source
This commit is contained in:
Cheng Wan
2025-06-04 15:53:22 -07:00
committed by GitHub
parent f0f84975f4
commit 81964328b7
22 changed files with 381 additions and 45 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
python/sglang/srt/layers/moe/topk.py
View File

@@ -303,6 +303,7 @@ def select_experts(
renormalize: bool,
topk_group: Optional[int] = None,
num_expert_group: Optional[int] = None,
num_fused_shared_experts: int = 0,
custom_routing_function: Optional[Callable] = None,
correction_bias: Optional[torch.Tensor] = None,
torch_native: bool = False,
@@ -310,7 +311,6 @@ def select_experts(
num_token_non_padded: Optional[torch.Tensor] = None,
expert_location_dispatch_info: Optional[ExpertLocationDispatchInfo] = None,
):
num_fused_shared_experts = global_server_args_dict["num_fused_shared_experts"]
router_logits, correction_bias = (
expert_location_dispatch.transform_select_experts_inputs(