phoenix-oss/llama-stack-mirror
0
0
Fork 1
mirror of https://github.com/meta-llama/llama-stack.git synced 2025-12-13 16:32:37 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

chore: Auto-detect Provider ID when only 1 Vector Store Provider available and Require Embedding Model

Browse source 
Signed-off-by: Francisco Javier Arceo <farceo@redhat.com>
This commit is contained in:
Francisco Javier Arceo 2025-10-13 10:09:04 -04:00
parent b95f095a54
commit 1bb3434562
3 changed files with 27 additions and 110 deletions
Download patch file Download diff file Expand all files Collapse all files

57
llama_stack/core/routers/vector_io.py
View file

@ -55,30 +55,18 @@ class VectorIORouter(VectorIO):
logger.debug("VectorIORouter.shutdown")
pass
async def _get_first_embedding_model(self) -> tuple[str, int] | None:
"""Get the first available embedding model identifier."""
try:
# Get all models from the routing table
all_models = await self.routing_table.get_all_with_type("model")
async def _get_embedding_model_dimension(self, embedding_model_id: str) -> int:
"""Get the embedding dimension for a specific embedding model."""
all_models = await self.routing_table.get_all_with_type("model")
# Filter for embedding models
embedding_models = [
model
for model in all_models
if hasattr(model, "model_type") and model.model_type == ModelType.embedding
]
if embedding_models:
dimension = embedding_models[0].metadata.get("embedding_dimension", None)
for model in all_models:
if model.identifier == embedding_model_id and model.model_type == ModelType.embedding:
dimension = model.metadata.get("embedding_dimension")
if dimension is None:
raise ValueError(f"Embedding model {embedding_models[0].identifier} has no embedding dimension")
return embedding_models[0].identifier, dimension
else:
logger.warning("No embedding models found in the routing table")
return None
except Exception as e:
logger.error(f"Error getting embedding models: {e}")
return None
raise ValueError(f"Embedding model '{embedding_model_id}' has no embedding_dimension in metadata")
return int(dimension)
raise ValueError(f"Embedding model '{embedding_model_id}' not found or not an embedding model")
async def register_vector_db(
self,
@ -134,15 +122,24 @@ class VectorIORouter(VectorIO):
logger.debug(f"VectorIORouter.openai_create_vector_store: name={params.name}, provider_id={provider_id}")
# If no embedding model is provided, use the first available one
# TODO: this branch will soon be deleted so you _must_ provide the embedding_model when
# creating a vector store
# Require explicit embedding model specification
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}")
raise ValueError("embedding_model is required in extra_body when creating a vector store")
# Always extract embedding dimension from the model registry
embedding_dimension = await self._get_embedding_model_dimension(embedding_model)
# Auto-select provider if not specified
if provider_id is None:
if len(self.routing_table.impls_by_provider_id) == 1:
provider_id = list(self.routing_table.impls_by_provider_id.keys())[0]
logger.info(f"No provider_id specified, using the only available vector_io provider: {provider_id}")
else:
available_providers = list(self.routing_table.impls_by_provider_id.keys())
raise ValueError(
f"Multiple vector_io providers available. Please specify provider_id in extra_body. "
f"Available providers: {available_providers}"
)
vector_db_id = f"vs_{uuid.uuid4()}"
registered_vector_db = await self.routing_table.register_vector_db(