Misc fix for min_p_sampling, --cuda-graph-bs (#2761)

2025-01-07 02:52:53 -08:00
parent 6d08ce2aa9
commit bdc1acf6cd
17 changed files with 135 additions and 63 deletions
--- a/python/sglang/srt/managers/scheduler.py
+++ b/python/sglang/srt/managers/scheduler.py
@@ -1516,8 +1516,9 @@ class Scheduler:
        return success, message

    def update_weights_from_distributed(
-        self, recv_req: UpdateWeightsFromDistributedReqInput
-    ):
+        self,
+        recv_req: UpdateWeightsFromDistributedReqInput,
+    ) -> Tuple[bool, str]:
        """Update the online model parameter."""
        success, message = self.tp_worker.update_weights_from_distributed(recv_req)
        if success: