Merge branch 'main' into content-extension

llama_stack/providers/utils/bedrock/config.py

@@ -4,53 +4,55 @@
 # This source code is licensed under the terms described in the LICENSE file in
 # the root directory of this source tree.
 
+import os
+
 from pydantic import BaseModel, Field
 
 
 class BedrockBaseConfig(BaseModel):
     aws_access_key_id: str | None = Field(
-        default=None,
+        default_factory=lambda: os.getenv("AWS_ACCESS_KEY_ID"),
         description="The AWS access key to use. Default use environment variable: AWS_ACCESS_KEY_ID",
     )
     aws_secret_access_key: str | None = Field(
-        default=None,
+        default_factory=lambda: os.getenv("AWS_SECRET_ACCESS_KEY"),
         description="The AWS secret access key to use. Default use environment variable: AWS_SECRET_ACCESS_KEY",
     )
     aws_session_token: str | None = Field(
-        default=None,
+        default_factory=lambda: os.getenv("AWS_SESSION_TOKEN"),
         description="The AWS session token to use. Default use environment variable: AWS_SESSION_TOKEN",
     )
     region_name: str | None = Field(
-        default=None,
+        default_factory=lambda: os.getenv("AWS_DEFAULT_REGION"),
         description="The default AWS Region to use, for example, us-west-1 or us-west-2."
         "Default use environment variable: AWS_DEFAULT_REGION",
     )
     profile_name: str | None = Field(
-        default=None,
+        default_factory=lambda: os.getenv("AWS_PROFILE"),
         description="The profile name that contains credentials to use.Default use environment variable: AWS_PROFILE",
     )
     total_max_attempts: int | None = Field(
-        default=None,
+        default_factory=lambda: int(val) if (val := os.getenv("AWS_MAX_ATTEMPTS")) else None,
         description="An integer representing the maximum number of attempts that will be made for a single request, "
         "including the initial attempt. Default use environment variable: AWS_MAX_ATTEMPTS",
     )
     retry_mode: str | None = Field(
-        default=None,
+        default_factory=lambda: os.getenv("AWS_RETRY_MODE"),
         description="A string representing the type of retries Boto3 will perform."
         "Default use environment variable: AWS_RETRY_MODE",
     )
     connect_timeout: float | None = Field(
-        default=60,
+        default_factory=lambda: float(os.getenv("AWS_CONNECT_TIMEOUT", "60")),
         description="The time in seconds till a timeout exception is thrown when attempting to make a connection. "
         "The default is 60 seconds.",
     )
     read_timeout: float | None = Field(
-        default=60,
+        default_factory=lambda: float(os.getenv("AWS_READ_TIMEOUT", "60")),
         description="The time in seconds till a timeout exception is thrown when attempting to read from a connection."
         "The default is 60 seconds.",
     )
     session_ttl: int | None = Field(
-        default=3600,
+        default_factory=lambda: int(os.getenv("AWS_SESSION_TTL", "3600")),
         description="The time in seconds till a session expires. The default is 3600 seconds (1 hour).",
     )
 

llama_stack/providers/utils/inference/embedding_mixin.py

@@ -4,6 +4,7 @@
 # This source code is licensed under the terms described in the LICENSE file in
 # the root directory of this source tree.
 
+import asyncio
 import base64
 import struct
 from typing import TYPE_CHECKING
@@ -43,9 +44,11 @@ class SentenceTransformerEmbeddingMixin:
         task_type: EmbeddingTaskType | None = None,
     ) -> EmbeddingsResponse:
         model = await self.model_store.get_model(model_id)
-        embedding_model = self._load_sentence_transformer_model(model.provider_resource_id)
-        embeddings = embedding_model.encode(
-            [interleaved_content_as_str(content) for content in contents], show_progress_bar=False
+        embedding_model = await self._load_sentence_transformer_model(model.provider_resource_id)
+        embeddings = await asyncio.to_thread(
+            embedding_model.encode,
+            [interleaved_content_as_str(content) for content in contents],
+            show_progress_bar=False,
         )
         return EmbeddingsResponse(embeddings=embeddings)
 
@@ -64,8 +67,8 @@ class SentenceTransformerEmbeddingMixin:
 
         # Get the model and generate embeddings
         model_obj = await self.model_store.get_model(model)
-        embedding_model = self._load_sentence_transformer_model(model_obj.provider_resource_id)
-        embeddings = embedding_model.encode(input_list, show_progress_bar=False)
+        embedding_model = await self._load_sentence_transformer_model(model_obj.provider_resource_id)
+        embeddings = await asyncio.to_thread(embedding_model.encode, input_list, show_progress_bar=False)
 
         # Convert embeddings to the requested format
         data = []
@@ -93,7 +96,7 @@ class SentenceTransformerEmbeddingMixin:
             usage=usage,
         )
 
-    def _load_sentence_transformer_model(self, model: str) -> "SentenceTransformer":
+    async def _load_sentence_transformer_model(self, model: str) -> "SentenceTransformer":
         global EMBEDDING_MODELS
 
         loaded_model = EMBEDDING_MODELS.get(model)
@@ -101,8 +104,12 @@ class SentenceTransformerEmbeddingMixin:
             return loaded_model
 
         log.info(f"Loading sentence transformer for {model}...")
-        from sentence_transformers import SentenceTransformer
 
-        loaded_model = SentenceTransformer(model)
+        def _load_model():
+            from sentence_transformers import SentenceTransformer
+
+            return SentenceTransformer(model)
+
+        loaded_model = await asyncio.to_thread(_load_model)
         EMBEDDING_MODELS[model] = loaded_model
         return loaded_model

llama_stack/providers/utils/memory/vector_store.py

@@ -294,12 +294,12 @@ class VectorDBWithIndex:
                 _validate_embedding(c.embedding, i, self.vector_db.embedding_dimension)
 
         if chunks_to_embed:
-            resp = await self.inference_api.embeddings(
+            resp = await self.inference_api.openai_embeddings(
                 self.vector_db.embedding_model,
                 [c.content for c in chunks_to_embed],
             )
-            for c, embedding in zip(chunks_to_embed, resp.embeddings, strict=False):
-                c.embedding = embedding
+            for c, data in zip(chunks_to_embed, resp.data, strict=False):
+                c.embedding = data.embedding
 
         embeddings = np.array([c.embedding for c in chunks], dtype=np.float32)
         await self.index.add_chunks(chunks, embeddings)
@@ -334,8 +334,8 @@ class VectorDBWithIndex:
         if mode == "keyword":
             return await self.index.query_keyword(query_string, k, score_threshold)
 
-        embeddings_response = await self.inference_api.embeddings(self.vector_db.embedding_model, [query_string])
-        query_vector = np.array(embeddings_response.embeddings[0], dtype=np.float32)
+        embeddings_response = await self.inference_api.openai_embeddings(self.vector_db.embedding_model, [query_string])
+        query_vector = np.array(embeddings_response.data[0].embedding, dtype=np.float32)
         if mode == "hybrid":
             return await self.index.query_hybrid(
                 query_vector, query_string, k, score_threshold, reranker_type, reranker_params

llama_stack/providers/utils/sqlstore/sqlalchemy_sqlstore.py

@@ -23,6 +23,7 @@ from sqlalchemy import (
 )
 from sqlalchemy.ext.asyncio import async_sessionmaker, create_async_engine
 from sqlalchemy.ext.asyncio.engine import AsyncEngine
+from sqlalchemy.sql.elements import ColumnElement
 
 from llama_stack.apis.common.responses import PaginatedResponse
 from llama_stack.log import get_logger
@@ -43,6 +44,30 @@ TYPE_MAPPING: dict[ColumnType, Any] = {
 }
 
 
+def _build_where_expr(column: ColumnElement, value: Any) -> ColumnElement:
+    """Return a SQLAlchemy expression for a where condition.
+
+    `value` may be a simple scalar (equality) or a mapping like {">": 123}.
+    The returned expression is a SQLAlchemy ColumnElement usable in query.where(...).
+    """
+    if isinstance(value, Mapping):
+        if len(value) != 1:
+            raise ValueError(f"Operator mapping must have a single operator, got: {value}")
+        op, operand = next(iter(value.items()))
+        if op == "==" or op == "=":
+            return column == operand
+        if op == ">":
+            return column > operand
+        if op == "<":
+            return column < operand
+        if op == ">=":
+            return column >= operand
+        if op == "<=":
+            return column <= operand
+        raise ValueError(f"Unsupported operator '{op}' in where mapping")
+    return column == value
+
+
 class SqlAlchemySqlStoreImpl(SqlStore):
     def __init__(self, config: SqlAlchemySqlStoreConfig):
         self.config = config
@@ -111,7 +136,7 @@ class SqlAlchemySqlStoreImpl(SqlStore):
 
             if where:
                 for key, value in where.items():
-                    query = query.where(table_obj.c[key] == value)
+                    query = query.where(_build_where_expr(table_obj.c[key], value))
 
             if where_sql:
                 query = query.where(text(where_sql))
@@ -222,7 +247,7 @@ class SqlAlchemySqlStoreImpl(SqlStore):
         async with self.async_session() as session:
             stmt = self.metadata.tables[table].update()
             for key, value in where.items():
-                stmt = stmt.where(self.metadata.tables[table].c[key] == value)
+                stmt = stmt.where(_build_where_expr(self.metadata.tables[table].c[key], value))
             await session.execute(stmt, data)
             await session.commit()
 
@@ -233,7 +258,7 @@ class SqlAlchemySqlStoreImpl(SqlStore):
         async with self.async_session() as session:
             stmt = self.metadata.tables[table].delete()
             for key, value in where.items():
-                stmt = stmt.where(self.metadata.tables[table].c[key] == value)
+                stmt = stmt.where(_build_where_expr(self.metadata.tables[table].c[key], value))
             await session.execute(stmt)
             await session.commit()
 

llama_stack/providers/utils/tools/mcp.py

@@ -67,6 +67,38 @@ async def client_wrapper(endpoint: str, headers: dict[str, str]) -> AsyncGenerat
                     raise AuthenticationRequiredError(exc) from exc
             if i == len(connection_strategies) - 1:
                 raise
+        except* httpx.ConnectError as eg:
+            # Connection refused, server down, network unreachable
+            if i == len(connection_strategies) - 1:
+                error_msg = f"Failed to connect to MCP server at {endpoint}: Connection refused"
+                logger.error(f"MCP connection error: {error_msg}")
+                raise ConnectionError(error_msg) from eg
+            else:
+                logger.warning(
+                    f"failed to connect to MCP server at {endpoint} via {strategy.name}, falling back to {connection_strategies[i + 1].name}"
+                )
+        except* httpx.TimeoutException as eg:
+            # Request timeout, server too slow
+            if i == len(connection_strategies) - 1:
+                error_msg = f"MCP server at {endpoint} timed out"
+                logger.error(f"MCP timeout error: {error_msg}")
+                raise TimeoutError(error_msg) from eg
+            else:
+                logger.warning(
+                    f"MCP server at {endpoint} timed out via {strategy.name}, falling back to {connection_strategies[i + 1].name}"
+                )
+        except* httpx.RequestError as eg:
+            # DNS resolution failures, network errors, invalid URLs
+            if i == len(connection_strategies) - 1:
+                # Get the first exception's message for the error string
+                exc_msg = str(eg.exceptions[0]) if eg.exceptions else "Unknown error"
+                error_msg = f"Network error connecting to MCP server at {endpoint}: {exc_msg}"
+                logger.error(f"MCP network error: {error_msg}")
+                raise ConnectionError(error_msg) from eg
+            else:
+                logger.warning(
+                    f"network error connecting to MCP server at {endpoint} via {strategy.name}, falling back to {connection_strategies[i + 1].name}"
+                )
         except* McpError:
             if i < len(connection_strategies) - 1:
                 logger.warning(

llama_stack/providers/utils/vector_io/vector_utils.py

@@ -37,3 +37,122 @@ def sanitize_collection_name(name: str, weaviate_format=False) -> str:
     else:
         s = proper_case(re.sub(r"[^a-zA-Z0-9]", "", name))
     return s
+
+
+class WeightedInMemoryAggregator:
+    @staticmethod
+    def _normalize_scores(scores: dict[str, float]) -> dict[str, float]:
+        """
+        Normalize scores to 0-1 range using min-max normalization.
+
+        Args:
+            scores: dictionary of scores with document IDs as keys and scores as values
+
+        Returns:
+            Normalized scores with document IDs as keys and normalized scores as values
+        """
+        if not scores:
+            return {}
+        min_score, max_score = min(scores.values()), max(scores.values())
+        score_range = max_score - min_score
+        if score_range > 0:
+            return {doc_id: (score - min_score) / score_range for doc_id, score in scores.items()}
+        return dict.fromkeys(scores, 1.0)
+
+    @staticmethod
+    def weighted_rerank(
+        vector_scores: dict[str, float],
+        keyword_scores: dict[str, float],
+        alpha: float = 0.5,
+    ) -> dict[str, float]:
+        """
+        Rerank via weighted average of scores.
+
+        Args:
+            vector_scores: scores from vector search
+            keyword_scores: scores from keyword search
+            alpha: weight factor between 0 and 1 (default: 0.5)
+                   0 = keyword only, 1 = vector only, 0.5 = equal weight
+
+        Returns:
+            All unique document IDs with weighted combined scores
+        """
+        all_ids = set(vector_scores.keys()) | set(keyword_scores.keys())
+        normalized_vector_scores = WeightedInMemoryAggregator._normalize_scores(vector_scores)
+        normalized_keyword_scores = WeightedInMemoryAggregator._normalize_scores(keyword_scores)
+
+        # Weighted formula: score = (1-alpha) * keyword_score + alpha * vector_score
+        # alpha=0 means keyword only, alpha=1 means vector only
+        return {
+            doc_id: ((1 - alpha) * normalized_keyword_scores.get(doc_id, 0.0))
+            + (alpha * normalized_vector_scores.get(doc_id, 0.0))
+            for doc_id in all_ids
+        }
+
+    @staticmethod
+    def rrf_rerank(
+        vector_scores: dict[str, float],
+        keyword_scores: dict[str, float],
+        impact_factor: float = 60.0,
+    ) -> dict[str, float]:
+        """
+        Rerank via Reciprocal Rank Fusion.
+
+        Args:
+            vector_scores: scores from vector search
+            keyword_scores: scores from keyword search
+            impact_factor: impact factor for RRF (default: 60.0)
+
+        Returns:
+            All unique document IDs with RRF combined scores
+        """
+
+        # Convert scores to ranks
+        vector_ranks = {
+            doc_id: i + 1
+            for i, (doc_id, _) in enumerate(sorted(vector_scores.items(), key=lambda x: x[1], reverse=True))
+        }
+        keyword_ranks = {
+            doc_id: i + 1
+            for i, (doc_id, _) in enumerate(sorted(keyword_scores.items(), key=lambda x: x[1], reverse=True))
+        }
+
+        all_ids = set(vector_scores.keys()) | set(keyword_scores.keys())
+        rrf_scores = {}
+        for doc_id in all_ids:
+            vector_rank = vector_ranks.get(doc_id, float("inf"))
+            keyword_rank = keyword_ranks.get(doc_id, float("inf"))
+
+            # RRF formula: score = 1/(k + r) where k is impact_factor (default: 60.0) and r is the rank
+            rrf_scores[doc_id] = (1.0 / (impact_factor + vector_rank)) + (1.0 / (impact_factor + keyword_rank))
+        return rrf_scores
+
+    @staticmethod
+    def combine_search_results(
+        vector_scores: dict[str, float],
+        keyword_scores: dict[str, float],
+        reranker_type: str = "rrf",
+        reranker_params: dict[str, float] | None = None,
+    ) -> dict[str, float]:
+        """
+        Combine vector and keyword search results using specified reranking strategy.
+
+        Args:
+            vector_scores: scores from vector search
+            keyword_scores: scores from keyword search
+            reranker_type: type of reranker to use (default: RERANKER_TYPE_RRF)
+            reranker_params: parameters for the reranker
+
+        Returns:
+            All unique document IDs with combined scores
+        """
+        if reranker_params is None:
+            reranker_params = {}
+
+        if reranker_type == "weighted":
+            alpha = reranker_params.get("alpha", 0.5)
+            return WeightedInMemoryAggregator.weighted_rerank(vector_scores, keyword_scores, alpha)
+        else:
+            # Default to RRF for None, RRF, or any unknown types
+            impact_factor = reranker_params.get("impact_factor", 60.0)
+            return WeightedInMemoryAggregator.rrf_rerank(vector_scores, keyword_scores, impact_factor)