[V1] LoRA - Enable Serving Usecase (#12883)

Signed-off-by: Varun Sundar Rabindranath <varun@neuralmagic.com>
Co-authored-by: Varun Sundar Rabindranath <varun@neuralmagic.com>

vllm/v1/engine/core.py

@@ -13,6 +13,7 @@ import zmq.asyncio
 
 from vllm.config import VllmConfig
 from vllm.logger import init_logger
+from vllm.lora.request import LoRARequest
 from vllm.transformers_utils.config import (
     maybe_register_config_serialize_by_value)
 from vllm.utils import get_exception_traceback, zmq_socket_ctx
@@ -146,6 +147,9 @@ class EngineCore:
     def reset_prefix_cache(self):
         self.scheduler.reset_prefix_cache()
 
+    def add_lora(self, lora_request: LoRARequest) -> None:
+        self.model_executor.add_lora(lora_request)
+
 
 class EngineCoreProc(EngineCore):
     """ZMQ-wrapper for running EngineCore in background process."""
@@ -262,12 +266,15 @@ class EngineCoreProc(EngineCore):
             self.reset_prefix_cache()
         elif request_type == EngineCoreRequestType.PROFILE:
             self.model_executor.profile(request)
+        elif request_type == EngineCoreRequestType.ADD_LORA:
+            self.model_executor.add_lora(request)
 
     def process_input_socket(self, input_path: str):
         """Input socket IO thread."""
 
         # Msgpack serialization decoding.
         add_request_decoder = MsgpackDecoder(EngineCoreRequest)
+        add_lora_decoder = MsgpackDecoder(LoRARequest)
         generic_decoder = MsgpackDecoder()
 
         with zmq_socket_ctx(input_path, zmq.constants.PULL) as socket:
@@ -277,9 +284,14 @@ class EngineCoreProc(EngineCore):
                 request_type = EngineCoreRequestType(bytes(type_frame.buffer))
 
                 # Deserialize the request data.
-                decoder = add_request_decoder if (
-                    request_type
-                    == EngineCoreRequestType.ADD) else generic_decoder
+                decoder = None
+                if request_type == EngineCoreRequestType.ADD:
+                    decoder = add_request_decoder
+                elif request_type == EngineCoreRequestType.ADD_LORA:
+                    decoder = add_lora_decoder
+                else:
+                    decoder = generic_decoder
+
                 request = decoder.decode(data_frame.buffer)
 
                 # Push to input queue for core busy loop.