Merge remote-tracking branch 'upstream/main' into update-api-docs

llama_stack/apis/common/training_types.py

@@ -27,7 +27,7 @@ class PostTrainingMetric(BaseModel):
     perplexity: float
 
 
-@json_schema_type(schema={"description": "Checkpoint created during training runs"})
+@json_schema_type
 class Checkpoint(BaseModel):
     """Checkpoint created during training runs.
 

llama_stack/apis/inference/inference.py

@@ -911,6 +911,12 @@ class OpenAIEmbeddingsResponse(BaseModel):
 class ModelStore(Protocol):
     async def get_model(self, identifier: str) -> Model: ...
 
+    async def update_registered_llm_models(
+        self,
+        provider_id: str,
+        models: list[Model],
+    ) -> None: ...
+
 
 class TextTruncation(Enum):
     """Config for how to truncate text for embedding when text is longer than the model's max sequence length. Start and End semantics depend on whether the language is left-to-right or right-to-left.

llama_stack/apis/models/models.py

@@ -7,7 +7,7 @@
 from enum import StrEnum
 from typing import Any, Literal, Protocol, runtime_checkable
 
-from pydantic import BaseModel, ConfigDict, Field
+from pydantic import BaseModel, ConfigDict, Field, field_validator
 
 from llama_stack.apis.resource import Resource, ResourceType
 from llama_stack.providers.utils.telemetry.trace_protocol import trace_protocol
@@ -51,13 +51,21 @@ class Model(CommonModelFields, Resource):
         return self.identifier
 
     @property
-    def provider_model_id(self) -> str | None:
+    def provider_model_id(self) -> str:
+        assert self.provider_resource_id is not None, "Provider resource ID must be set"
         return self.provider_resource_id
 
     model_config = ConfigDict(protected_namespaces=())
 
     model_type: ModelType = Field(default=ModelType.llm)
 
+    @field_validator("provider_resource_id")
+    @classmethod
+    def validate_provider_resource_id(cls, v):
+        if v is None:
+            raise ValueError("provider_resource_id cannot be None")
+        return v
+
 
 class ModelInput(CommonModelFields):
     model_id: str

llama_stack/apis/post_training/post_training.py

@@ -181,6 +181,14 @@ class RLHFAlgorithm(Enum):
     dpo = "dpo"
 
 
+@json_schema_type
+class DPOLossType(Enum):
+    sigmoid = "sigmoid"
+    hinge = "hinge"
+    ipo = "ipo"
+    kto_pair = "kto_pair"
+
+
 @json_schema_type
 class DPOAlignmentConfig(BaseModel):
     """Configuration for Direct Preference Optimization (DPO) alignment.
@@ -189,12 +197,16 @@ class DPOAlignmentConfig(BaseModel):
     :param reward_clip: Maximum absolute value for reward clipping
     :param epsilon: Small value added for numerical stability
     :param gamma: Discount factor for future rewards
+    :param beta: Temperature parameter for the DPO loss
+    :param loss_type: The type of loss function to use for DPO
     """
 
     reward_scale: float
     reward_clip: float
     epsilon: float
     gamma: float
+    beta: float
+    loss_type: DPOLossType = DPOLossType.sigmoid
 
 
 @json_schema_type

llama_stack/apis/telemetry/telemetry.py

@@ -164,7 +164,7 @@ class MetricInResponse(BaseModel):
 
 # This is a short term solution to allow inference API to return metrics
 # The ideal way to do this is to have a way for all response types to include metrics
-# and all metric events logged to the telemetry API to be inlcuded with the response
+# and all metric events logged to the telemetry API to be included with the response
 # To do this, we will need to augment all response types with a metrics field.
 # We have hit a blocker from stainless SDK that prevents us from doing this.
 # The blocker is that if we were to augment the response types that have a data field

llama_stack/apis/tools/rag_tool.py

@@ -4,7 +4,7 @@
 # This source code is licensed under the terms described in the LICENSE file in
 # the root directory of this source tree.
 
-from enum import Enum
+from enum import Enum, StrEnum
 from typing import Annotated, Any, Literal, Protocol
 
 from pydantic import BaseModel, Field, field_validator
@@ -100,6 +100,20 @@ class RAGQueryGenerator(Enum):
     custom = "custom"
 
 
+@json_schema_type
+class RAGSearchMode(StrEnum):
+    """
+    Search modes for RAG query retrieval:
+    - VECTOR: Uses vector similarity search for semantic matching
+    - KEYWORD: Uses keyword-based search for exact matching
+    - HYBRID: Combines both vector and keyword search for better results
+    """
+
+    VECTOR = "vector"
+    KEYWORD = "keyword"
+    HYBRID = "hybrid"
+
+
 @json_schema_type
 class DefaultRAGQueryGeneratorConfig(BaseModel):
     """Configuration for the default RAG query generator.
@@ -154,7 +168,7 @@ class RAGQueryConfig(BaseModel):
     max_tokens_in_context: int = 4096
     max_chunks: int = 5
     chunk_template: str = "Result {index}\nContent: {chunk.content}\nMetadata: {metadata}\n"
-    mode: str | None = None
+    mode: RAGSearchMode | None = RAGSearchMode.VECTOR
     ranker: Ranker | None = Field(default=None)  # Only used for hybrid mode
 
     @field_validator("chunk_template")

llama_stack/apis/vector_dbs/vector_dbs.py

@@ -26,6 +26,7 @@ class VectorDB(Resource):
 
     embedding_model: str
     embedding_dimension: int
+    vector_db_name: str | None = None
 
     @property
     def vector_db_id(self) -> str:
@@ -48,6 +49,7 @@ class VectorDBInput(BaseModel):
     vector_db_id: str
     embedding_model: str
     embedding_dimension: int
+    provider_id: str | None = None
     provider_vector_db_id: str | None = None
 
 
@@ -90,6 +92,7 @@ class VectorDBs(Protocol):
         embedding_model: str,
         embedding_dimension: int | None = 384,
         provider_id: str | None = None,
+        vector_db_name: str | None = None,
         provider_vector_db_id: str | None = None,
     ) -> VectorDB:
         """Register a vector database.
@@ -98,6 +101,7 @@ class VectorDBs(Protocol):
         :param embedding_model: The embedding model to use.
         :param embedding_dimension: The dimension of the embedding model.
         :param provider_id: The identifier of the provider.
+        :param vector_db_name: The name of the vector database.
         :param provider_vector_db_id: The identifier of the vector database in the provider.
         :returns: A VectorDB.
         """

llama_stack/apis/vector_io/vector_io.py

@@ -475,7 +475,7 @@ class VectorIO(Protocol):
     @webmethod(route="/openai/v1/vector_stores", method="POST")
     async def openai_create_vector_store(
         self,
-        name: str,
+        name: str | None = None,
         file_ids: list[str] | None = None,
         expires_after: dict[str, Any] | None = None,
         chunking_strategy: dict[str, Any] | None = None,
@@ -483,7 +483,6 @@ class VectorIO(Protocol):
         embedding_model: str | None = None,
         embedding_dimension: int | None = 384,
         provider_id: str | None = None,
-        provider_vector_db_id: str | None = None,
     ) -> VectorStoreObject:
         """Creates a vector store.
 
@@ -495,7 +494,6 @@ class VectorIO(Protocol):
         :param embedding_model: The embedding model to use for this vector store.
         :param embedding_dimension: The dimension of the embedding vectors (default: 384).
         :param provider_id: The ID of the provider to use for this vector store.
-        :param provider_vector_db_id: The provider-specific vector database ID.
         :returns: A VectorStoreObject representing the created vector store.
         """
         ...