Merge vl execution path into normal execution path (#2829)

* merge vl model into gpu_model runner

Change-Id: I9f4691a3d5f135e8d72b1d58abcd15ef3aa3f2a6

* fix chinese

Change-Id: Ic7405109b984c21e076fb3b01ff6feb571d0119a

* fix the parse parameter

Change-Id: I4cd62ee87c06220af580d91e347145d4394917fe

* fix the bug in online_inference

Change-Id: Idb111bb2114e83017c4050b2a68cf039c6d3c559

* polish code

Change-Id: I7d4194102c2f1b0743b74fbd5fc284eb8ef4d17c

fastdeploy/worker/worker_process.py

@@ -549,6 +549,10 @@ def parse_args():
         "'ipc_snapshot': load from disk snapshot of IPC weights, "
         "'meta': provide RL traing worker, no_weights_load"
         "'normal':normal load weight")
+    parser.add_argument("--enable_mm",
+                        type=str,
+                        default="false",
+                        help="Whether to use vl")
     parser.add_argument("--enable_logprob",
                         action='store_true',
                         help="Enable output of token-level log probabilities.")
@@ -650,6 +654,8 @@ def initialize_fd_config(args, ranks: int = 1, local_rank: int = 0) -> FDConfig:
             "No quantization config found and use original weight and act dtype."
         )
 
+    # Set VL tag
+    model_config.enable_mm = getattr(args, 'enable_mm', 'false').lower() == 'true'
     logger.info(f"- Dynamic load weight: {load_config.dynamic_load_weight}")
     logger.info(f"- Load strategy: {load_config.load_strategy}")