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

2025-06-04 15:53:22 -07:00
parent f0f84975f4
commit 81964328b7
22 changed files with 381 additions and 45 deletions
--- a/python/sglang/srt/layers/quantization/init.py
+++ b/python/sglang/srt/layers/quantization/init.py
@@ -289,6 +289,7 @@ def monkey_patch_moe_apply(class_obj: "FusedMoEMethodBase"):
        use_grouped_topk: 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,
        activation: str = "silu",