Merge 84a26339c8 into 61582f327c

.github/actions/setup-test-environment/action.yml

@@ -39,6 +39,17 @@ runs:
       if: ${{ inputs.provider == 'vllm' && inputs.inference-mode == 'record' }}
       uses: ./.github/actions/setup-vllm
 
+    - name: Set provider URLs for replay mode
+      if: ${{ inputs.inference-mode == 'replay' }}
+      shell: bash
+      run: |
+        # setting so providers get registered in replay mode
+        if [ "${{ inputs.provider }}" == "ollama" ]; then
+          echo "OLLAMA_URL=http://localhost:11434" >> $GITHUB_ENV
+        elif [ "${{ inputs.provider }}" == "vllm" ]; then
+          echo "VLLM_URL=http://localhost:8000/v1" >> $GITHUB_ENV
+        fi
+
     - name: Build Llama Stack
       shell: bash
       run: |

docs/_static/llama-stack-spec.html

@@ -14997,6 +14997,47 @@
                     "text": {
                         "type": "string",
                         "description": "The actual text content"
+                    },
+                    "embedding": {
+                        "type": "array",
+                        "items": {
+                            "type": "number"
+                        },
+                        "description": "(Optional) Embedding vector for the content, if available"
+                    },
+                    "created_timestamp": {
+                        "type": "integer",
+                        "description": "(Optional) Timestamp when the content was created"
+                    },
+                    "metadata": {
+                        "type": "object",
+                        "additionalProperties": {
+                            "oneOf": [
+                                {
+                                    "type": "null"
+                                },
+                                {
+                                    "type": "boolean"
+                                },
+                                {
+                                    "type": "number"
+                                },
+                                {
+                                    "type": "string"
+                                },
+                                {
+                                    "type": "array"
+                                },
+                                {
+                                    "type": "object"
+                                }
+                            ]
+                        },
+                        "description": "(Optional) Metadata associated with the content, such as source, author, etc."
+                    },
+                    "chunk_metadata": {
+                        "$ref": "#/components/schemas/ChunkMetadata",
+                        "description": "(Optional) Metadata associated with the chunk, such as document ID, source, etc."
                     }
                 },
                 "additionalProperties": false,

docs/_static/llama-stack-spec.yaml

@@ -11143,6 +11143,34 @@ components:
         text:
           type: string
           description: The actual text content
+        embedding:
+          type: array
+          items:
+            type: number
+          description: >-
+            (Optional) Embedding vector for the content, if available
+        created_timestamp:
+          type: integer
+          description: >-
+            (Optional) Timestamp when the content was created
+        metadata:
+          type: object
+          additionalProperties:
+            oneOf:
+              - type: 'null'
+              - type: boolean
+              - type: number
+              - type: string
+              - type: array
+              - type: object
+          description: >-
+            (Optional) Metadata associated with the content, such as source, author,
+            etc.
+        chunk_metadata:
+          $ref: '#/components/schemas/ChunkMetadata'
+          description: >-
+            (Optional) Metadata associated with the chunk, such as document ID, source,
+            etc.
       additionalProperties: false
       required:
         - type

llama_stack/apis/vector_io/vector_io.py

@@ -226,10 +226,18 @@ class VectorStoreContent(BaseModel):
 
     :param type: Content type, currently only "text" is supported
     :param text: The actual text content
+    :param embedding: (Optional) Embedding vector for the content, if available
+    :param created_timestamp: (Optional) Timestamp when the content was created
+    :param metadata: (Optional) Metadata associated with the content, such as source, author, etc.
+    :param chunk_metadata: (Optional) Metadata associated with the chunk, such as document ID, source, etc.
     """
 
     type: Literal["text"]
     text: str
+    embedding: list[float] | None = None
+    created_timestamp: int | None = None
+    metadata: dict[str, Any] | None = None
+    chunk_metadata: ChunkMetadata | None = None
 
 
 @json_schema_type

llama_stack/providers/utils/memory/openai_vector_store_mixin.py

@@ -17,6 +17,7 @@ from llama_stack.apis.files import Files, OpenAIFileObject
 from llama_stack.apis.vector_dbs import VectorDB
 from llama_stack.apis.vector_io import (
     Chunk,
+    ChunkMetadata,
     QueryChunksResponse,
     SearchRankingOptions,
     VectorStoreChunkingStrategy,
@@ -520,31 +521,68 @@ class OpenAIVectorStoreMixin(ABC):
             raise ValueError(f"Unsupported filter type: {filter_type}")
 
     def _chunk_to_vector_store_content(self, chunk: Chunk) -> list[VectorStoreContent]:
+        created_ts = None
+        if chunk.chunk_metadata is not None:
+            created_ts = getattr(chunk.chunk_metadata, "created_timestamp", None)
+
+        metadata_dict = {}
+        if chunk.chunk_metadata:
+            if hasattr(chunk.chunk_metadata, "model_dump"):
+                metadata_dict = chunk.chunk_metadata.model_dump()
+            else:
+                metadata_dict = vars(chunk.chunk_metadata)
+
+        user_metadata = chunk.metadata or {}
+        base_meta = {**metadata_dict, **user_metadata}
+
         # content is InterleavedContent
         if isinstance(chunk.content, str):
             content = [
                 VectorStoreContent(
                     type="text",
                     text=chunk.content,
+                    embedding=chunk.embedding,
+                    created_timestamp=created_ts,
+                    metadata=user_metadata,
+                    chunk_metadata=ChunkMetadata(**base_meta) if base_meta else None,
                 )
             ]
         elif isinstance(chunk.content, list):
             # TODO: Add support for other types of content
-            content = [
+            content = []
+            for item in chunk.content:
+                if hasattr(item, "type") and item.type == "text":
+                    item_meta = {**base_meta}
+                    item_user_meta = getattr(item, "metadata", {}) or {}
+                    if item_user_meta:
+                        item_meta.update(item_user_meta)
+
+                    content.append(
                         VectorStoreContent(
                             type="text",
                             text=item.text,
+                            embedding=getattr(item, "embedding", None),
+                            created_timestamp=created_ts,
+                            metadata=item_user_meta,
+                            chunk_metadata=ChunkMetadata(**item_meta) if item_meta else None,
+                        )
                     )
-                for item in chunk.content
-                if item.type == "text"
-            ]
         else:
-            if chunk.content.type != "text":
-                raise ValueError(f"Unsupported content type: {chunk.content.type}")
+            content_item = chunk.content
+            if content_item.type != "text":
+                raise ValueError(f"Unsupported content type: {content_item.type}")
+
+            item_user_meta = getattr(content_item, "metadata", {}) or {}
+            combined_meta = {**base_meta, **item_user_meta}
+
             content = [
                 VectorStoreContent(
                     type="text",
-                    text=chunk.content.text,
+                    text=content_item.text,
+                    embedding=getattr(content_item, "embedding", None),
+                    created_timestamp=created_ts,
+                    metadata=item_user_meta,
+                    chunk_metadata=ChunkMetadata(**combined_meta) if combined_meta else None,
                 )
             ]
         return content

tests/integration/README.md

@@ -108,9 +108,7 @@ pytest -s -v tests/integration/inference/ \
 Running Vector IO tests for a number of embedding models:
 
 ```bash
-EMBEDDING_MODELS=all-MiniLM-L6-v2
-
-pytest -s -v tests/integration/vector_io/ \
-   --stack-config=inference=sentence-transformers,vector_io=sqlite-vec \
-   --embedding-model=$EMBEDDING_MODELS
+uv run pytest -sv --stack-config="inference=inline::sentence-transformers,vector_io=inline::sqlite-vec,files=localfs" \
+tests/integration/vector_io --embedding-model \
+sentence-transformers/all-MiniLM-L6-v2
 ```

tests/integration/vector_io/test_openai_vector_stores.py

@@ -6,6 +6,7 @@
 
 import logging
 import time
+import uuid
 from io import BytesIO
 
 import pytest
@@ -897,3 +898,76 @@ def test_openai_vector_store_search_modes(llama_stack_client, client_with_models
         search_mode=search_mode,
     )
     assert search_response is not None
+
+
+def test_openai_vector_store_file_contents_with_extended_fields(compat_client_with_empty_stores, client_with_models):
+    skip_if_provider_doesnt_support_openai_vector_stores(client_with_models)
+
+    compat_client = compat_client_with_empty_stores
+    vector_store = compat_client.vector_stores.create(
+        name="extended_fields_test_store", metadata={"purpose": "extended_fields_testing"}
+    )
+
+    test_content = b"This is a test document."
+    file_name = f"extended_fields_test_{uuid.uuid4().hex}.txt"
+    attributes = {"test_type": "extended_fields", "version": "1.0"}
+
+    with BytesIO(test_content) as file_buffer:
+        file_buffer.name = file_name
+        file = compat_client.files.create(file=file_buffer, purpose="assistants")
+
+    file_attach_response = compat_client.vector_stores.files.create(
+        vector_store_id=vector_store.id,
+        file_id=file.id,
+        attributes=attributes,
+    )
+
+    assert file_attach_response.status == "completed", f"File attach failed: {file_attach_response.last_error}"
+    assert file_attach_response.attributes == attributes
+
+    file_contents = compat_client.vector_stores.files.content(
+        vector_store_id=vector_store.id,
+        file_id=file.id,
+    )
+
+    assert file_contents
+    assert file_contents.filename == file_name
+    assert file_contents.attributes == attributes
+    assert len(file_contents.content) > 0
+
+    for content_item in file_contents.content:
+        if isinstance(compat_client, LlamaStackClient):
+            content_item = content_item.to_dict()
+        assert content_item["type"] == "text"
+        assert "text" in content_item
+        assert isinstance(content_item["text"], str)
+        assert len(content_item["text"]) > 0
+
+        if "embedding" in content_item:
+            assert isinstance(content_item["embedding"], list)
+            assert all(isinstance(x, (int | float)) for x in content_item["embedding"])
+
+        if "created_timestamp" in content_item:
+            assert isinstance(content_item["created_timestamp"], int)
+            assert content_item["created_timestamp"] > 0
+
+        if "chunk_metadata" in content_item:
+            assert isinstance(content_item["chunk_metadata"], dict)
+            if "chunk_id" in content_item["chunk_metadata"]:
+                assert isinstance(content_item["chunk_metadata"]["chunk_id"], str)
+            if "chunk_window" in content_item["chunk_metadata"]:
+                assert isinstance(content_item["chunk_metadata"]["chunk_window"], str)
+
+    search_response = compat_client.vector_stores.search(
+        vector_store_id=vector_store.id, query="test document", max_num_results=5
+    )
+
+    assert search_response is not None
+    assert len(search_response.data) > 0
+
+    for result_object in search_response.data:
+        result = result_object.to_dict()
+        assert "content" in result
+        assert len(result["content"]) > 0
+        assert result["content"][0]["type"] == "text"
+        assert "text" in result["content"][0]

tests/unit/providers/vector_io/test_vector_io_openai_vector_stores.py

@@ -12,7 +12,7 @@ import numpy as np
 import pytest
 
 from llama_stack.apis.vector_dbs import VectorDB
-from llama_stack.apis.vector_io import Chunk, QueryChunksResponse
+from llama_stack.apis.vector_io import Chunk, ChunkMetadata, QueryChunksResponse, VectorStoreContent
 from llama_stack.providers.remote.vector_io.milvus.milvus import VECTOR_DBS_PREFIX
 
 # This test is a unit test for the inline VectoerIO providers. This should only contain
@@ -294,3 +294,35 @@ async def test_delete_openai_vector_store_file_from_storage(vector_io_adapter, t
     assert loaded_file_info == {}
     loaded_contents = await vector_io_adapter._load_openai_vector_store_file_contents(store_id, file_id)
     assert loaded_contents == []
+
+
+async def test_chunk_to_vector_store_content_with_new_fields(vector_io_adapter):
+    sample_chunk_metadata = ChunkMetadata(
+        chunk_id="chunk123",
+        document_id="doc456",
+        source="test_source",
+        created_timestamp=1625133600,
+        updated_timestamp=1625133600,
+        chunk_window="0-100",
+        chunk_tokenizer="test_tokenizer",
+        chunk_embedding_model="dummy_model",
+        chunk_embedding_dimension=384,
+        content_token_count=100,
+        metadata_token_count=100,
+    )
+
+    sample_chunk = Chunk(
+        content="hello world", metadata={"lang": "en"}, embedding=[0.5, 0.7, 0.9], chunk_metadata=sample_chunk_metadata
+    )
+
+    vsc_list: VectorStoreContent = vector_io_adapter._chunk_to_vector_store_content(sample_chunk)
+    assert isinstance(vsc_list, list)
+    assert len(vsc_list) > 0
+
+    vsc = vsc_list[0]
+    assert vsc.text == "hello world"
+    assert vsc.type == "text"
+    assert vsc.metadata == {"lang": "en"}
+    assert vsc.chunk_metadata == sample_chunk_metadata
+    assert vsc.embedding == [0.5, 0.7, 0.9]
+    assert vsc.created_timestamp == 1625133600