phoenix-oss/llama-stack-mirror
0
0
Fork 1
mirror of https://github.com/meta-llama/llama-stack.git synced 2025-10-10 13:28:40 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Merge 354ed48598 into a7f9ce9a3a

Browse source
This commit is contained in:
Francisco Arceo 2025-09-23 12:27:14 +01:00 • committed by GitHub
commit b3d377cc35
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
8 changed files with 247 additions and 15 deletions
Download patch file Download diff file Expand all files Collapse all files

8
llama_stack/apis/vector_io/vector_io.py
View file

@ -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

60
llama_stack/providers/utils/memory/openai_vector_store_mixin.py
View file

@ -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 = [
VectorStoreContent(
type="text",
text=item.text,
)
for item in chunk.content
if item.type == "text"
]
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,
)
)
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