phoenix-oss/llama-stack-mirror
0
0
Fork 1
mirror of https://github.com/meta-llama/llama-stack.git synced 2025-12-23 00:52:26 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

feat(vector-io): configurable embedding models for all providers (v2)\n\nAdds embedding_model and embedding_dimension fields to all VectorIOConfig classes.\nRouter respects provider defaults with fallback.\nIntroduces embedding_utils helper.\nComprehensive docs & samples.\nResolves #2729

Browse source
This commit is contained in:
skamenan7 2025-07-17 11:51:40 -04:00
parent c8f274347d
commit d55dd3e9a0
24 changed files with 482 additions and 14 deletions
Download patch file Download diff file Expand all files Collapse all files

31
llama_stack/distribution/routers/vector_io.py
View file

@ -129,13 +129,32 @@ class VectorIORouter(VectorIO):
) -> VectorStoreObject:
logger.debug(f"VectorIORouter.openai_create_vector_store: name={name}, provider_id={provider_id}")
# If no embedding model is provided, use the first available one
# If no embedding model is provided, try provider defaults then fallback
if embedding_model is None:
embedding_model_info = await self._get_first_embedding_model()
if embedding_model_info is None:
raise ValueError("No embedding model provided and no embedding models available in the system")
embedding_model, embedding_dimension = embedding_model_info
logger.info(f"No embedding model specified, using first available: {embedding_model}")
# Try to get provider-specific embedding model configuration
if provider_id:
try:
provider_impl = self.routing_table.get_provider_impl(provider_id)
provider_config = getattr(provider_impl, "config", None)
if provider_config:
if hasattr(provider_config, "embedding_model") and provider_config.embedding_model:
embedding_model = provider_config.embedding_model
logger.info(f"Using provider config default embedding model: {embedding_model}")
if hasattr(provider_config, "embedding_dimension") and provider_config.embedding_dimension:
embedding_dimension = provider_config.embedding_dimension
logger.info(f"Using provider config embedding dimension: {embedding_dimension}")
except Exception as e:
logger.debug(f"Could not get provider config for {provider_id}: {e}")
# If still no embedding model, use system fallback
if embedding_model is None:
embedding_model_info = await self._get_first_embedding_model()
if embedding_model_info is None:
raise ValueError("No embedding model provided and no embedding models available in the system")
embedding_model, embedding_dimension = embedding_model_info
logger.info(f"No embedding model specified, using first available: {embedding_model}")
vector_db_id = f"vs_{uuid.uuid4()}"
registered_vector_db = await self.routing_table.register_vector_db(