support non disturbing remote instance weight loader v2 (#14997)

Signed-off-by: Anqi Shen <amy.saq@antgroup.com>

python/sglang/srt/managers/scheduler.py

@@ -2656,6 +2656,9 @@ class Scheduler(
         self.send_to_detokenizer.send_output(recv_req, recv_req)
         return None
 
+    def get_remote_instance_transfer_engine_info(self):
+        return self.tp_worker.get_remote_instance_transfer_engine_info()
+
 
 class IdleSleeper:
     """
@@ -2769,14 +2772,25 @@ def run_scheduler_process(
             pp_rank,
             dp_rank,
         )
-        pipe_writer.send(
-            {
-                "status": "ready",
-                "max_total_num_tokens": scheduler.max_total_num_tokens,
-                "max_req_input_len": scheduler.max_req_input_len,
-            }
-        )
+        result_dict = {
+            "status": "ready",
+            "max_total_num_tokens": scheduler.max_total_num_tokens,
+            "max_req_input_len": scheduler.max_req_input_len,
+        }
+        if server_args.remote_instance_weight_loader_use_transfer_engine():
+            (
+                remote_instance_transfer_engine_session_id,
+                remote_instance_transfer_engine_weights_info_dict,
+            ) = scheduler.get_remote_instance_transfer_engine_info()
+            result_dict.update(
+                {
+                    "tp_rank": tp_rank,
+                    "remote_instance_transfer_engine_session_id": remote_instance_transfer_engine_session_id,
+                    "remote_instance_transfer_engine_weights_info_dict": remote_instance_transfer_engine_weights_info_dict,
+                }
+            )
 
+        pipe_writer.send(result_dict)
         disaggregation_mode: DisaggregationMode = scheduler.disaggregation_mode
         if disaggregation_mode == DisaggregationMode.NULL:
             if scheduler.enable_pdmux:

python/sglang/srt/managers/tp_worker.py

@@ -366,6 +366,12 @@ class TpModelWorker(BaseTpWorker):
             can_run_cuda_graph=can_run_cuda_graph,
         )
 
+    def get_remote_instance_transfer_engine_info(self):
+        return (
+            self.model_runner.remote_instance_transfer_engine_session_id,
+            self.model_runner.remote_instance_transfer_engine_weight_info,
+        )
+
     def forward_batch_generation(
         self,
         model_worker_batch: ModelWorkerBatch,