[feat] update bucketed weights from distributed (#13824)

Co-authored-by: Stefan He <hebiaobuaa@gmail.com>

python/sglang/srt/managers/io_struct.py

@@ -1144,6 +1144,8 @@ class UpdateWeightsFromDistributedReqInput(BaseReq):
     abort_all_requests: bool = False
     # Optional: Update weight version along with weights
     weight_version: Optional[str] = None
+    # Optional format specification for loading
+    load_format: Optional[str] = None
 
 
 @dataclass

python/sglang/srt/managers/tp_worker.py

@@ -153,7 +153,11 @@ class BaseTpWorker(ABC):
         self, recv_req: UpdateWeightsFromDistributedReqInput
     ):
         success, message = self.model_runner.update_weights_from_distributed(
-            recv_req.names, recv_req.dtypes, recv_req.shapes, recv_req.group_name
+            recv_req.names,
+            recv_req.dtypes,
+            recv_req.shapes,
+            recv_req.group_name,
+            recv_req.load_format,
         )
         return success, message