Throwing an exception when the model does not support pool tasks

Signed-off-by: zxw <1020938856@qq.com>

Author: yyzxw

vllm/v1/worker/gpu_model_runner.py

@@ -3375,8 +3375,17 @@ def _dummy_pooler_run(
         hidden_states: torch.Tensor,
     ) -> PoolerOutput:
         # Find the task that has the largest output for subsequent steps
+        supported_pooling_tasks = self.get_supported_pooling_tasks()
+
+        if not supported_pooling_tasks:
+            raise RuntimeError(
+                f"Model {self.model_config.model} does not support "
+                "any pooling tasks. see "
+                "https://docs.vllm.ai/en/latest/models/pooling_models.html "
+                "to learn more.")
+
         output_size = dict[PoolingTask, float]()
-        for task in self.get_supported_pooling_tasks():
+        for task in supported_pooling_tasks:
             # Run a full batch with each task to ensure none of them OOMs
             output = self._dummy_pooler_run_task(hidden_states, task)
             output_size[task] = sum(o.nbytes for o in output)