[BugFix] Fix Embedding Models with TP>1 (vllm-project#5075)

vllm/worker/embedding_model_runner.py

@@ -79,6 +79,10 @@ def execute_model(
             execute_model_kwargs.update({"image_input": multi_modal_input})
         hidden_states = model_executable(**execute_model_kwargs)
 
+        # Only perform pooling in the driver worker.
+        if not self.is_driver_worker:
+            return None
+
         return self.model.pooler(hidden_states=hidden_states,
                                  pooling_metadata=pooling_metadata)