[Feat] Support update weights without restart server (#1157)

python/sglang/srt/server.py

@@ -51,7 +51,11 @@ from sglang.srt.managers.controller_single import (
     start_controller_process as start_controller_process_single,
 )
 from sglang.srt.managers.detokenizer_manager import start_detokenizer_process
-from sglang.srt.managers.io_struct import EmbeddingReqInput, GenerateReqInput
+from sglang.srt.managers.io_struct import (
+    EmbeddingReqInput,
+    GenerateReqInput,
+    UpdateWeightReqInput,
+)
 from sglang.srt.managers.tokenizer_manager import TokenizerManager
 from sglang.srt.openai_api.adapter import (
     load_chat_template_for_openai_api,
@@ -136,6 +140,23 @@ async def flush_cache():
     )
 
 
+@app.post("/update_weights")
+async def update_weights(obj: UpdateWeightReqInput, request: Request):
+
+    success, message = await tokenizer_manager.update_weights(obj, request)
+    content = {"message": message, "success": str(success)}
+    if success:
+        return JSONResponse(
+            content,
+            status_code=HTTPStatus.OK,
+        )
+    else:
+        return JSONResponse(
+            content,
+            status_code=HTTPStatus.BAD_REQUEST,
+        )
+
+
 async def generate_request(obj: GenerateReqInput, request: Request):
     """Handle a generate request."""
     if obj.stream: