[Model] Explicit interface for vLLM models and support OOT embedding models (#9108)

tests/models/test_oot_registration.py

@@ -2,7 +2,7 @@ import os
 
 import pytest
 
-from vllm import LLM, SamplingParams
+from vllm import LLM, PoolingParams, SamplingParams
 from vllm.assets.image import ImageAsset
 
 from ..utils import fork_new_process_for_each_test
@@ -17,7 +17,7 @@ def test_plugin(dummy_opt_path):
 
 
 @fork_new_process_for_each_test
-def test_oot_registration(dummy_opt_path):
+def test_oot_registration_text_generation(dummy_opt_path):
     os.environ["VLLM_PLUGINS"] = "register_dummy_model"
     prompts = ["Hello, my name is", "The text does not matter"]
     sampling_params = SamplingParams(temperature=0)
@@ -32,11 +32,23 @@ def test_oot_registration(dummy_opt_path):
         assert rest == ""
 
 
+@fork_new_process_for_each_test
+def test_oot_registration_embedding(dummy_gemma2_embedding_path):
+    os.environ["VLLM_PLUGINS"] = "register_dummy_model"
+    prompts = ["Hello, my name is", "The text does not matter"]
+    sampling_params = PoolingParams()
+    llm = LLM(model=dummy_gemma2_embedding_path, load_format="dummy")
+    outputs = llm.encode(prompts, sampling_params)
+
+    for output in outputs:
+        assert all(v == 0 for v in output.outputs.embedding)
+
+
 image = ImageAsset("cherry_blossom").pil_image.convert("RGB")
 
 
 @fork_new_process_for_each_test
-def test_oot_multimodal_registration(dummy_llava_path):
+def test_oot_registration_multimodal(dummy_llava_path):
     os.environ["VLLM_PLUGINS"] = "register_dummy_model"
     prompts = [{
         "prompt": "What's in the image?<image>",