[Misc] Collect model support info in a single process per model (vllm…

…-project#9233)

Signed-off-by: Sumit Dubey <sumit.dubey2@ibm.com>

docs/source/models/adding_model.rst

@@ -99,7 +99,7 @@ This method should load the weights from the HuggingFace's checkpoint file and a
 5. Register your model
 ----------------------
 
-Finally, register your :code:`*ForCausalLM` class to the :code:`_MODELS` in `vllm/model_executor/models/registry.py <https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/models/registry.py>`_.
+Finally, register your :code:`*ForCausalLM` class to the :code:`_VLLM_MODELS` in `vllm/model_executor/models/registry.py <https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/models/registry.py>`_.
 
 6. Out-of-Tree Model Integration
 --------------------------------------------

vllm/engine/arg_utils.py

@@ -183,6 +183,8 @@ class EngineArgs:
     def __post_init__(self):
         if self.tokenizer is None:
             self.tokenizer = self.model
+
+        # Setup plugins
         from vllm.plugins import load_general_plugins
         load_general_plugins()
 

vllm/engine/multiprocessing/engine.py

@@ -130,6 +130,9 @@ def dead_error(self) -> BaseException:
     def from_engine_args(cls, engine_args: AsyncEngineArgs,
                          usage_context: UsageContext, ipc_path: str):
         """Creates an MQLLMEngine from the engine arguments."""
+        # Setup plugins for each process
+        from vllm.plugins import load_general_plugins
+        load_general_plugins()
 
         engine_config = engine_args.create_engine_config()