Refactor TopK to ensure readability and extensibility (#9338)

2025-09-14 19:16:25 -07:00
parent b7d385e812
commit 4844fac91d
14 changed files with 52 additions and 47 deletions
--- a/python/sglang/srt/models/gpt_oss.py
+++ b/python/sglang/srt/models/gpt_oss.py
@@ -121,7 +121,7 @@ class GptOssSparseMoeBlock(nn.Module):
        )

        self.top_k = config.num_experts_per_tok
-        experts_type = get_moe_impl_class()
+        experts_type = get_moe_impl_class(quant_config)
        extra_kwargs = {}
        if experts_type.__name__ == "FusedMoE":
            quant_config_name = (