Revert several PRs (#14958)

Co-authored-by: fzyzcjy <ch271828n@outlook.com>

python/sglang/srt/entrypoints/engine.py

@@ -127,18 +127,13 @@ class Engine(EngineBase):
         atexit.register(self.shutdown)
 
         # Launch subprocesses
-        (
-            tokenizer_manager,
-            template_manager,
-            scheduler_info,
-            port_args,
-            remote_instance_transfer_engine_info,
-        ) = _launch_subprocesses(server_args=server_args)
+        tokenizer_manager, template_manager, scheduler_info, port_args = (
+            _launch_subprocesses(server_args=server_args)
+        )
         self.tokenizer_manager = tokenizer_manager
         self.template_manager = template_manager
         self.scheduler_info = scheduler_info
         self.port_args = port_args
-        self.remote_instance_transfer_engine_info = remote_instance_transfer_engine_info
 
         # Initialize ZMQ sockets
         context = zmq.Context(2)
@@ -915,7 +910,6 @@ def _launch_subprocesses(
 
     # Wait for the model to finish loading
     scheduler_infos = []
-    remote_instance_transfer_engine_info = {}
     for i in range(len(scheduler_pipe_readers)):
         try:
             data = scheduler_pipe_readers[i].recv()
@@ -932,24 +926,9 @@ def _launch_subprocesses(
                 "Initialization failed. Please see the error messages above."
             )
         scheduler_infos.append(data)
-        if (
-            "tp_rank" in data
-            and "remote_instance_transfer_engine_session_id" in data
-            and "remote_instance_transfer_engine_weights_info_dict" in data
-        ):
-            remote_instance_transfer_engine_info[data["tp_rank"]] = (
-                data["remote_instance_transfer_engine_session_id"],
-                data["remote_instance_transfer_engine_weights_info_dict"],
-            )
 
     # Assume all schedulers have the same scheduler_info
     scheduler_info = scheduler_infos[0]
     tokenizer_manager.max_req_input_len = scheduler_info["max_req_input_len"]
 
-    return (
-        tokenizer_manager,
-        template_manager,
-        scheduler_info,
-        port_args,
-        remote_instance_transfer_engine_info,
-    )
+    return tokenizer_manager, template_manager, scheduler_info, port_args

python/sglang/srt/entrypoints/http_server.py

@@ -144,15 +144,6 @@ class _GlobalState:
     tokenizer_manager: Union[TokenizerManager, MultiTokenizerRouter, TokenizerWorker]
     template_manager: TemplateManager
     scheduler_info: Dict
-    # Dict{
-    #   rank: Tuple(
-    #           session_id,
-    #           Dict{
-    #               name: Tuple (d_ptr, numel, element_size)
-    #           }
-    #         )
-    # }
-    remote_instance_transfer_engine_info: Optional[Dict] = None
 
 
 _global_state: Optional[_GlobalState] = None
@@ -822,24 +813,6 @@ async def send_weights_to_remote_instance(
         return ORJSONResponse(content, status_code=HTTPStatus.BAD_REQUEST)
 
 
-@app.get("/get_remote_instance_transfer_engine_info")
-async def get_remote_instance_transfer_engine_info(rank: int = None):
-    if rank is None or rank < 0:
-        return Response(status_code=HTTPStatus.BAD_REQUEST)
-
-    try:
-        result = {
-            "rank": rank,
-            "remote_instance_transfer_engine_info": _global_state.remote_instance_transfer_engine_info[
-                rank
-            ],
-        }
-        return result
-    except Exception as e:
-        logger.error(f"Exception: {e}")
-        return Response(status_code=HTTPStatus.BAD_REQUEST)
-
-
 @app.post("/init_weights_update_group")
 async def init_weights_update_group(
     obj: InitWeightsUpdateGroupReqInput, request: Request
@@ -1413,20 +1386,15 @@ def launch_server(
     1. The HTTP server, Engine, and TokenizerManager all run in the main process.
     2. Inter-process communication is done through IPC (each process uses a different port) via the ZMQ library.
     """
-    (
-        tokenizer_manager,
-        template_manager,
-        scheduler_info,
-        port_args,
-        remote_instance_transfer_engine_info,
-    ) = _launch_subprocesses(server_args=server_args)
+    tokenizer_manager, template_manager, scheduler_info, port_args = (
+        _launch_subprocesses(server_args=server_args)
+    )
 
     set_global_state(
         _GlobalState(
             tokenizer_manager=tokenizer_manager,
             template_manager=template_manager,
             scheduler_info=scheduler_info,
-            remote_instance_transfer_engine_info=remote_instance_transfer_engine_info,
         )
     )