feat: add static embedding metadata to dynamic model listings for providers using OpenAIMixin (#3547)

# What does this PR do?

- remove auto-download of ollama embedding models
- add embedding model metadata to dynamic listing w/ unit test
- add support and tests for allowed_models
- removed inference provider models.py files where dynamic listing is
enabled
- store embedding metadata in embedding_model_metadata field on
inference providers
- make model_entries optional on ModelRegistryHelper and
LiteLLMOpenAIMixin
- make OpenAIMixin a ModelRegistryHelper
- skip base64 embedding test for remote::ollama, always returns floats
- only use OpenAI client for ollama model listing
- remove unused build_model_entry function
- remove unused get_huggingface_repo function


## Test Plan

ci w/ new tests

tests/integration/recordings/responses/models-bd032f995f2a-3255f444.json

@@ -0,0 +1,96 @@
+{
+  "request": {
+    "method": "POST",
+    "url": "http://0.0.0.0:11434/v1/v1/models",
+    "headers": {},
+    "body": {},
+    "endpoint": "/v1/models",
+    "model": ""
+  },
+  "response": {
+    "body": [
+      {
+        "__type__": "openai.types.model.Model",
+        "__data__": {
+          "id": "qwen3:8b",
+          "created": 1758707188,
+          "object": "model",
+          "owned_by": "library"
+        }
+      },
+      {
+        "__type__": "openai.types.model.Model",
+        "__data__": {
+          "id": "nomic-embed-text:137m-v1.5-fp16",
+          "created": 1758640855,
+          "object": "model",
+          "owned_by": "library"
+        }
+      },
+      {
+        "__type__": "openai.types.model.Model",
+        "__data__": {
+          "id": "nomic-embed-text:latest",
+          "created": 1756727155,
+          "object": "model",
+          "owned_by": "library"
+        }
+      },
+      {
+        "__type__": "openai.types.model.Model",
+        "__data__": {
+          "id": "llama3.2-vision:11b",
+          "created": 1756722893,
+          "object": "model",
+          "owned_by": "library"
+        }
+      },
+      {
+        "__type__": "openai.types.model.Model",
+        "__data__": {
+          "id": "llama-guard3:1b",
+          "created": 1756671473,
+          "object": "model",
+          "owned_by": "library"
+        }
+      },
+      {
+        "__type__": "openai.types.model.Model",
+        "__data__": {
+          "id": "all-minilm:l6-v2",
+          "created": 1756655274,
+          "object": "model",
+          "owned_by": "library"
+        }
+      },
+      {
+        "__type__": "openai.types.model.Model",
+        "__data__": {
+          "id": "all-minilm:latest",
+          "created": 1747317111,
+          "object": "model",
+          "owned_by": "library"
+        }
+      },
+      {
+        "__type__": "openai.types.model.Model",
+        "__data__": {
+          "id": "llama3.2:3b-instruct-fp16",
+          "created": 1744974677,
+          "object": "model",
+          "owned_by": "library"
+        }
+      },
+      {
+        "__type__": "openai.types.model.Model",
+        "__data__": {
+          "id": "llama3.2:3b",
+          "created": 1743536220,
+          "object": "model",
+          "owned_by": "library"
+        }
+      }
+    ],
+    "is_streaming": false
+  }
+}

tests/unit/providers/inference/test_litellm_openai_mixin.py

@@ -26,7 +26,6 @@ class TestProviderDataValidator(BaseModel):
 class TestLiteLLMAdapter(LiteLLMOpenAIMixin):
     def __init__(self, config: TestConfig):
         super().__init__(
-            model_entries=[],
             litellm_provider_name="test",
             api_key_from_config=config.api_key,
             provider_data_api_key_field="test_api_key",

tests/unit/providers/nvidia/test_eval.py

@@ -150,7 +150,7 @@ class TestNVIDIAEvalImpl(unittest.TestCase):
         self._assert_request_body(
             {
                 "config": f"nvidia/{MOCK_BENCHMARK_ID}",
-                "target": {"type": "model", "model": "meta/llama-3.1-8b-instruct"},
+                "target": {"type": "model", "model": "Llama3.1-8B-Instruct"},
             }
         )
 

tests/unit/providers/utils/inference/test_openai_mixin.py

@@ -13,7 +13,6 @@ from llama_stack.apis.models import ModelType
 from llama_stack.providers.utils.inference.openai_mixin import OpenAIMixin
 
 
-# Test implementation of OpenAIMixin for testing purposes
 class OpenAIMixinImpl(OpenAIMixin):
     def __init__(self):
         self.__provider_id__ = "test-provider"
@@ -25,12 +24,35 @@ class OpenAIMixinImpl(OpenAIMixin):
         raise NotImplementedError("This method should be mocked in tests")
 
 
+class OpenAIMixinWithEmbeddingsImpl(OpenAIMixin):
+    """Test implementation with embedding model metadata"""
+
+    embedding_model_metadata = {
+        "text-embedding-3-small": {"embedding_dimension": 1536, "context_length": 8192},
+        "text-embedding-ada-002": {"embedding_dimension": 1536, "context_length": 8192},
+    }
+
+    __provider_id__ = "test-provider"
+
+    def get_api_key(self) -> str:
+        raise NotImplementedError("This method should be mocked in tests")
+
+    def get_base_url(self) -> str:
+        raise NotImplementedError("This method should be mocked in tests")
+
+
 @pytest.fixture
 def mixin():
     """Create a test instance of OpenAIMixin"""
     return OpenAIMixinImpl()
 
 
+@pytest.fixture
+def mixin_with_embeddings():
+    """Create a test instance of OpenAIMixin with embedding model metadata"""
+    return OpenAIMixinWithEmbeddingsImpl()
+
+
 @pytest.fixture
 def mock_models():
     """Create multiple mock OpenAI model objects"""
@@ -181,3 +203,89 @@ class TestOpenAIMixinCacheBehavior:
             assert "some-mock-model-id" in mixin._model_cache
             assert "another-mock-model-id" in mixin._model_cache
             assert "final-mock-model-id" in mixin._model_cache
+
+
+class TestOpenAIMixinEmbeddingModelMetadata:
+    """Test cases for embedding_model_metadata attribute functionality"""
+
+    async def test_embedding_model_identified_and_augmented(self, mixin_with_embeddings, mock_client_context):
+        """Test that models in embedding_model_metadata are correctly identified as embeddings with metadata"""
+        # Create mock models: 1 embedding model and 1 LLM, while there are 2 known embedding models
+        mock_embedding_model = MagicMock(id="text-embedding-3-small")
+        mock_llm_model = MagicMock(id="gpt-4")
+        mock_models = [mock_embedding_model, mock_llm_model]
+
+        mock_client = MagicMock()
+
+        async def mock_models_list():
+            for model in mock_models:
+                yield model
+
+        mock_client.models.list.return_value = mock_models_list()
+
+        with mock_client_context(mixin_with_embeddings, mock_client):
+            result = await mixin_with_embeddings.list_models()
+
+            assert result is not None
+            assert len(result) == 2
+
+            # Find the models in the result
+            embedding_model = next(m for m in result if m.identifier == "text-embedding-3-small")
+            llm_model = next(m for m in result if m.identifier == "gpt-4")
+
+            # Check embedding model
+            assert embedding_model.model_type == ModelType.embedding
+            assert embedding_model.metadata == {"embedding_dimension": 1536, "context_length": 8192}
+            assert embedding_model.provider_id == "test-provider"
+            assert embedding_model.provider_resource_id == "text-embedding-3-small"
+
+            # Check LLM model
+            assert llm_model.model_type == ModelType.llm
+            assert llm_model.metadata == {}  # No metadata for LLMs
+            assert llm_model.provider_id == "test-provider"
+            assert llm_model.provider_resource_id == "gpt-4"
+
+
+class TestOpenAIMixinAllowedModels:
+    """Test cases for allowed_models filtering functionality"""
+
+    async def test_list_models_with_allowed_models_filter(self, mixin, mock_client_with_models, mock_client_context):
+        """Test that list_models filters models based on allowed_models set"""
+        mixin.allowed_models = {"some-mock-model-id", "another-mock-model-id"}
+
+        with mock_client_context(mixin, mock_client_with_models):
+            result = await mixin.list_models()
+
+            assert result is not None
+            assert len(result) == 2
+
+            model_ids = [model.identifier for model in result]
+            assert "some-mock-model-id" in model_ids
+            assert "another-mock-model-id" in model_ids
+            assert "final-mock-model-id" not in model_ids
+
+    async def test_list_models_with_empty_allowed_models(self, mixin, mock_client_with_models, mock_client_context):
+        """Test that empty allowed_models set allows all models"""
+        assert len(mixin.allowed_models) == 0
+
+        with mock_client_context(mixin, mock_client_with_models):
+            result = await mixin.list_models()
+
+            assert result is not None
+            assert len(result) == 3  # All models should be included
+
+            model_ids = [model.identifier for model in result]
+            assert "some-mock-model-id" in model_ids
+            assert "another-mock-model-id" in model_ids
+            assert "final-mock-model-id" in model_ids
+
+    async def test_check_model_availability_with_allowed_models(
+        self, mixin, mock_client_with_models, mock_client_context
+    ):
+        """Test that check_model_availability respects allowed_models"""
+        mixin.allowed_models = {"final-mock-model-id"}
+
+        with mock_client_context(mixin, mock_client_with_models):
+            assert await mixin.check_model_availability("final-mock-model-id")
+            assert not await mixin.check_model_availability("some-mock-model-id")
+            assert not await mixin.check_model_availability("another-mock-model-id")

tests/unit/providers/utils/test_model_registry.py

@@ -84,14 +84,14 @@ def unknown_model() -> Model:
 
 @pytest.fixture
 def helper(known_provider_model: ProviderModelEntry, known_provider_model2: ProviderModelEntry) -> ModelRegistryHelper:
-    return ModelRegistryHelper([known_provider_model, known_provider_model2])
+    return ModelRegistryHelper(model_entries=[known_provider_model, known_provider_model2])
 
 
 class MockModelRegistryHelperWithDynamicModels(ModelRegistryHelper):
     """Test helper that simulates a provider with dynamically available models."""
 
     def __init__(self, model_entries: list[ProviderModelEntry], available_models: list[str]):
-        super().__init__(model_entries)
+        super().__init__(model_entries=model_entries)
         self._available_models = available_models
 
     async def check_model_availability(self, model: str) -> bool: