[Frontend][4/n] Improve pooling entrypoints | pooling. (#39153)

Signed-off-by: wang.yuqi <yuqi.wang@daocloud.io>

vllm/plugins/io_processors/__init__.py

@@ -12,6 +12,23 @@ from vllm.utils.import_utils import resolve_obj_by_qualname
 logger = logging.getLogger(__name__)
 
 
+def has_io_processor(
+    vllm_config: VllmConfig,
+    plugin_from_init: str | None = None,
+):
+    if plugin_from_init:
+        model_plugin = plugin_from_init
+    else:
+        # A plugin can be specified via the model config
+        # Retrieve the model specific plugin if available
+        # This is using a custom field in the hf_config for the model
+        hf_config = vllm_config.model_config.hf_config.to_dict()
+        config_plugin = hf_config.get("io_processor_plugin")
+        model_plugin = config_plugin
+
+    return model_plugin is not None
+
+
 def get_io_processor(
     vllm_config: VllmConfig,
     renderer: BaseRenderer,

vllm/v1/engine/async_llm.py

@@ -26,7 +26,6 @@ from vllm.logger import init_logger
 from vllm.lora.request import LoRARequest
 from vllm.multimodal import MULTIMODAL_REGISTRY, MultiModalRegistry
 from vllm.outputs import STREAM_FINISHED, PoolingRequestOutput, RequestOutput
-from vllm.plugins.io_processors import get_io_processor
 from vllm.pooling_params import PoolingParams
 from vllm.renderers import renderer_from_config
 from vllm.renderers.inputs.preprocess import extract_prompt_components
@@ -133,11 +132,6 @@ class AsyncLLM(EngineClient):
             )
 
         self.renderer = renderer = renderer_from_config(self.vllm_config)
-        self.io_processor = get_io_processor(
-            self.vllm_config,
-            self.renderer,
-            self.model_config.io_processor_plugin,
-        )
 
         # Convert EngineInput --> EngineCoreRequest.
         self.input_processor = InputProcessor(self.vllm_config, renderer)

vllm/v1/engine/llm_engine.py

@@ -19,7 +19,6 @@ from vllm.logger import init_logger
 from vllm.lora.request import LoRARequest
 from vllm.multimodal import MULTIMODAL_REGISTRY, MultiModalRegistry
 from vllm.outputs import PoolingRequestOutput, RequestOutput
-from vllm.plugins.io_processors import get_io_processor
 from vllm.pooling_params import PoolingParams
 from vllm.renderers import renderer_from_config
 from vllm.renderers.inputs.preprocess import extract_prompt_components
@@ -90,11 +89,6 @@ class LLMEngine:
         self.should_execute_dummy_batch = False
 
         self.renderer = renderer = renderer_from_config(self.vllm_config)
-        self.io_processor = get_io_processor(
-            self.vllm_config,
-            self.renderer,
-            self.model_config.io_processor_plugin,
-        )
 
         # Convert EngineInput --> EngineCoreRequest.
         self.input_processor = InputProcessor(self.vllm_config, renderer)