almost there

vllm/entrypoints/openai/rpc/__init__.py

@@ -31,9 +31,10 @@ class RPCUtilityRequest(Enum):
     GET_MODEL_CONFIG = 2
     GET_DECODING_CONFIG = 3
     GET_PARALLEL_CONFIG = 4
-    GET_LORA_CONFIG = 5
-    DO_LOG_STATS = 6
-    CHECK_HEALTH = 7
+    GET_SCHEDULER_CONFIG = 5
+    GET_LORA_CONFIG = 6
+    DO_LOG_STATS = 7
+    CHECK_HEALTH = 8
 
 
 RPC_REQUEST_TYPE = Union[RPCGenerateRequest, RPCAbortRequest,

vllm/entrypoints/openai/rpc/client.py

@@ -90,8 +90,7 @@ async def _send_one_way_rpc_request(self, request: RPC_REQUEST_TYPE,
         return response
 
     async def get_tokenizer(self, lora_request: LoRARequest):
-        # TODO: handle this via get data? - or avoid doing via RPC
-        return self.tokenizer
+        await self.tokenizer.get_lora_tokenizer_async(lora_request)
 
     async def is_tracing_enabled(self):
         # TODO: what is this?
@@ -144,12 +143,11 @@ async def get_lora_config(self):
         """Get LoRAConfig from the RPCServer"""
 
         return await self._send_get_data_rpc_request(
-            RPCUtilityRequest.GET_SCHEDULER_CONFIG,
+            RPCUtilityRequest.GET_LORA_CONFIG,
             expected_type=LoRAConfig,
             error_message="Could not get LoRAConfig from RPC Server"
         )
 
-
     async def abort(self, request_id: str):
         """Send an ABORT_REQUEST signal to the RPC Server"""
 

vllm/entrypoints/openai/rpc/server.py

@@ -154,6 +154,8 @@ def _make_handler_coro(self, identity,
                 return self.get_parallel_config(identity)
             elif request == RPCUtilityRequest.GET_DECODING_CONFIG:
                 return self.get_decoding_config(identity)
+            elif request == RPCUtilityRequest.GET_SCHEDULER_CONFIG:
+                return self.get_scheduler_config(identity)
             elif request == RPCUtilityRequest.GET_LORA_CONFIG:
                 return self.get_lora_config(identity)
             elif request == RPCUtilityRequest.DO_LOG_STATS: