Support sanity checking weight consistency especially for RL (#13854)

python/sglang/srt/entrypoints/http_server.py

@@ -76,6 +76,7 @@ from sglang.srt.environ import envs
 from sglang.srt.function_call.function_call_parser import FunctionCallParser
 from sglang.srt.managers.io_struct import (
     AbortReq,
+    CheckWeightsReqInput,
     CloseSessionReqInput,
     ConfigureLoggingReq,
     ContinueGenerationReqInput,
@@ -956,6 +957,15 @@ async def resume_memory_occupation(
         return _create_error_response(e)
 
 
+@app.post("/weights_checker")
+async def check_weights(obj: CheckWeightsReqInput, request: Request):
+    success, message = await _global_state.tokenizer_manager.check_weights(obj, request)
+    return ORJSONResponse(
+        {"success": success, "message": message},
+        status_code=200 if success else HTTPStatus.BAD_REQUEST,
+    )
+
+
 @app.api_route("/slow_down", methods=["GET", "POST"])
 async def slow_down(obj: SlowDownReqInput, request: Request):
     """Slow down the system deliberately. Only for testing. Example scenario: