Fix bug of refactoring TopKOutput in w4afp8 (#8745)

Co-authored-by: luoyuan.luo <luoyuan.luo@antgroup.com>
2025-08-04 11:05:02 +08:00
parent f024795e57
commit 3b87a9e8ae
1 changed files with 1 additions and 1 deletions
--- a/python/sglang/srt/layers/quantization/w4afp8.py
+++ b/python/sglang/srt/layers/quantization/w4afp8.py
@@ -282,7 +282,7 @@ class W4AFp8MoEMethod(FusedMoEMethodBase):
        # TODO(ch-wan): move it out of this class
        from sglang.srt.layers.moe.cutlass_w4a8_moe import cutlass_w4a8_moe

-        topk_ids, topk_weights, _ = topk_output
+        topk_weights, topk_ids, _ = topk_output
        local_topk_ids = topk_ids
        if layer.expert_map is not None:
            "Translate info from expert_map to topk_ids"