Make embedding generation go through inference (#606)

This PR does the following:
1) adds the ability to generate embeddings in all supported inference
providers.
2) Moves all the memory providers to use the inference API and improved
the memory tests to setup the inference stack correctly and use the
embedding models

This is a merge from #589 and #598

llama_stack/providers/tests/inference/conftest.py

@@ -18,6 +18,12 @@ def pytest_addoption(parser):
         default=None,
         help="Specify the inference model to use for testing",
     )
+    parser.addoption(
+        "--embedding-model",
+        action="store",
+        default=None,
+        help="Specify the embedding model to use for testing",
+    )
 
 
 def pytest_configure(config):

llama_stack/providers/tests/inference/fixtures.py

@@ -9,9 +9,9 @@ import os
 import pytest
 import pytest_asyncio
 
-from llama_stack.apis.models import ModelInput
-
+from llama_stack.apis.models import ModelInput, ModelType
 from llama_stack.distribution.datatypes import Api, Provider
+
 from llama_stack.providers.inline.inference.meta_reference import (
     MetaReferenceInferenceConfig,
 )
@@ -47,6 +47,9 @@ def inference_meta_reference(inference_model) -> ProviderFixture:
     inference_model = (
         [inference_model] if isinstance(inference_model, str) else inference_model
     )
+    # If embedding dimension is set, use the 8B model for testing
+    if os.getenv("EMBEDDING_DIMENSION"):
+        inference_model = ["meta-llama/Llama-3.1-8B-Instruct"]
 
     return ProviderFixture(
         providers=[
@@ -85,7 +88,7 @@ def inference_ollama(inference_model) -> ProviderFixture:
     inference_model = (
         [inference_model] if isinstance(inference_model, str) else inference_model
     )
-    if "Llama3.1-8B-Instruct" in inference_model:
+    if inference_model and "Llama3.1-8B-Instruct" in inference_model:
         pytest.skip("Ollama only supports Llama3.2-3B-Instruct for testing")
 
     return ProviderFixture(
@@ -232,11 +235,23 @@ INFERENCE_FIXTURES = [
 async def inference_stack(request, inference_model):
     fixture_name = request.param
     inference_fixture = request.getfixturevalue(f"inference_{fixture_name}")
+    model_type = ModelType.llm
+    metadata = {}
+    if os.getenv("EMBEDDING_DIMENSION"):
+        model_type = ModelType.embedding_model
+        metadata["embedding_dimension"] = get_env_or_fail("EMBEDDING_DIMENSION")
+
     test_stack = await construct_stack_for_test(
         [Api.inference],
         {"inference": inference_fixture.providers},
         inference_fixture.provider_data,
-        models=[ModelInput(model_id=inference_model)],
+        models=[
+            ModelInput(
+                model_id=inference_model,
+                model_type=model_type,
+                metadata=metadata,
+            )
+        ],
     )
 
     return test_stack.impls[Api.inference], test_stack.impls[Api.models]

llama_stack/providers/tests/inference/test_embeddings.py

@@ -0,0 +1,62 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+
+import pytest
+
+from llama_stack.apis.inference import EmbeddingsResponse, ModelType
+
+# How to run this test:
+# pytest -v -s llama_stack/providers/tests/inference/test_embeddings.py
+
+
+class TestEmbeddings:
+    @pytest.mark.asyncio
+    async def test_embeddings(self, inference_model, inference_stack):
+        inference_impl, models_impl = inference_stack
+        model = await models_impl.get_model(inference_model)
+
+        if model.model_type != ModelType.embedding_model:
+            pytest.skip("This test is only applicable for embedding models")
+
+        response = await inference_impl.embeddings(
+            model_id=inference_model,
+            contents=["Hello, world!"],
+        )
+        assert isinstance(response, EmbeddingsResponse)
+        assert len(response.embeddings) > 0
+        assert all(isinstance(embedding, list) for embedding in response.embeddings)
+        assert all(
+            isinstance(value, float)
+            for embedding in response.embeddings
+            for value in embedding
+        )
+
+    @pytest.mark.asyncio
+    async def test_batch_embeddings(self, inference_model, inference_stack):
+        inference_impl, models_impl = inference_stack
+        model = await models_impl.get_model(inference_model)
+
+        if model.model_type != ModelType.embedding_model:
+            pytest.skip("This test is only applicable for embedding models")
+
+        texts = ["Hello, world!", "This is a test", "Testing embeddings"]
+
+        response = await inference_impl.embeddings(
+            model_id=inference_model,
+            contents=texts,
+        )
+
+        assert isinstance(response, EmbeddingsResponse)
+        assert len(response.embeddings) == len(texts)
+        assert all(isinstance(embedding, list) for embedding in response.embeddings)
+        assert all(
+            isinstance(value, float)
+            for embedding in response.embeddings
+            for value in embedding
+        )
+
+        embedding_dim = len(response.embeddings[0])
+        assert all(len(embedding) == embedding_dim for embedding in response.embeddings)