EPLB: prefer to use physical experts in the same gpu or node (#10874)

2025-10-29 12:01:11 +08:00
parent e03b6beeb1
commit ec92b0cefe
2 changed files with 111 additions and 37 deletions
--- a/python/sglang/srt/model_executor/model_runner.py
+++ b/python/sglang/srt/model_executor/model_runner.py
@@ -348,7 +348,11 @@ class ModelRunner:

        if not self.is_draft_worker:
            set_global_expert_location_metadata(
-                compute_initial_expert_location_metadata(server_args, self.model_config)
+                compute_initial_expert_location_metadata(
+                    server_args=server_args,
+                    model_config=self.model_config,
+                    moe_ep_rank=self.moe_ep_rank,
+                )
            )
            if self.tp_rank == 0 and get_bool_env_var(
                "SGLANG_LOG_EXPERT_LOCATION_METADATA"