phoenix-oss/llama-stack-mirror
0
0
Fork 1
mirror of https://github.com/meta-llama/llama-stack.git synced 2025-12-20 10:28:41 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Merge branch 'main' into add-batches

Browse source
This commit is contained in:
Matthew Farrellee 2025-08-13 07:33:41 -04:00
commit 95a3ecdffc
67 changed files with 1158 additions and 424 deletions
Download patch file Download diff file Expand all files Collapse all files

13
llama_stack/providers/inline/agents/meta_reference/agents.py
View file

@ -327,10 +327,21 @@ class MetaReferenceAgentsImpl(Agents):
temperature: float | None = None,
text: OpenAIResponseText | None = None,
tools: list[OpenAIResponseInputTool] | None = None,
include: list[str] | None = None,
max_infer_iters: int | None = 10,
) -> OpenAIResponseObject:
return await self.openai_responses_impl.create_openai_response(
input, model, instructions, previous_response_id, store, stream, temperature, text, tools, max_infer_iters
input,
model,
instructions,
previous_response_id,
store,
stream,
temperature,
text,
tools,
include,
max_infer_iters,
)
async def list_openai_responses(

19
llama_stack/providers/inline/agents/meta_reference/openai_responses.py
View file

@ -38,6 +38,7 @@ from llama_stack.apis.agents.openai_responses import (
OpenAIResponseOutputMessageContent,
OpenAIResponseOutputMessageContentOutputText,
OpenAIResponseOutputMessageFileSearchToolCall,
OpenAIResponseOutputMessageFileSearchToolCallResults,
OpenAIResponseOutputMessageFunctionToolCall,
OpenAIResponseOutputMessageMCPListTools,
OpenAIResponseOutputMessageWebSearchToolCall,
@ -333,6 +334,7 @@ class OpenAIResponsesImpl:
temperature: float | None = None,
text: OpenAIResponseText | None = None,
tools: list[OpenAIResponseInputTool] | None = None,
include: list[str] | None = None,
max_infer_iters: int | None = 10,
):
stream = bool(stream)
@ -486,8 +488,12 @@ class OpenAIResponsesImpl:
# Convert collected chunks to complete response
if chat_response_tool_calls:
tool_calls = [chat_response_tool_calls[i] for i in sorted(chat_response_tool_calls.keys())]
# when there are tool calls, we need to clear the content
chat_response_content = []
else:
tool_calls = None
assistant_message = OpenAIAssistantMessageParam(
content="".join(chat_response_content),
tool_calls=tool_calls,
@ -826,12 +832,13 @@ class OpenAIResponsesImpl:
text = result.metadata["chunks"][i] if "chunks" in result.metadata else None
score = result.metadata["scores"][i] if "scores" in result.metadata else None
message.results.append(
{
"file_id": doc_id,
"filename": doc_id,
"text": text,
"score": score,
}
OpenAIResponseOutputMessageFileSearchToolCallResults(
file_id=doc_id,
filename=doc_id,
text=text,
score=score,
attributes={},
)
)
if error_exc or (result.error_code and result.error_code > 0) or result.error_message:
message.status = "failed"

14
llama_stack/providers/inline/safety/prompt_guard/prompt_guard.py
View file

@ -15,6 +15,7 @@ from llama_stack.apis.safety import (
RunShieldResponse,
Safety,
SafetyViolation,
ShieldStore,
ViolationLevel,
)
from llama_stack.apis.shields import Shield
@ -32,6 +33,8 @@ PROMPT_GUARD_MODEL = "Prompt-Guard-86M"
class PromptGuardSafetyImpl(Safety, ShieldsProtocolPrivate):
shield_store: ShieldStore
def __init__(self, config: PromptGuardConfig, _deps) -> None:
self.config = config
@ -53,7 +56,7 @@ class PromptGuardSafetyImpl(Safety, ShieldsProtocolPrivate):
self,
shield_id: str,
messages: list[Message],
params: dict[str, Any] = None,
params: dict[str, Any],
) -> RunShieldResponse:
shield = await self.shield_store.get_shield(shield_id)
if not shield:
@ -61,6 +64,9 @@ class PromptGuardSafetyImpl(Safety, ShieldsProtocolPrivate):
return await self.shield.run(messages)
async def run_moderation(self, input: str | list[str], model: str):
raise NotImplementedError("run_moderation not implemented for PromptGuard")
class PromptGuardShield:
def __init__(
@ -117,8 +123,10 @@ class PromptGuardShield:
elif self.config.guard_type == PromptGuardType.jailbreak.value and score_malicious > self.threshold:
violation = SafetyViolation(
violation_level=ViolationLevel.ERROR,
violation_type=f"prompt_injection:malicious={score_malicious}",
violation_return_message="Sorry, I cannot do this.",
user_message="Sorry, I cannot do this.",
metadata={
"violation_type": f"prompt_injection:malicious={score_malicious}",
},
)
return RunShieldResponse(violation=violation)

27
llama_stack/providers/inline/vector_io/faiss/faiss.py
View file

@ -33,6 +33,7 @@ from llama_stack.providers.utils.kvstore import kvstore_impl
from llama_stack.providers.utils.kvstore.api import KVStore
from llama_stack.providers.utils.memory.openai_vector_store_mixin import OpenAIVectorStoreMixin
from llama_stack.providers.utils.memory.vector_store import (
ChunkForDeletion,
EmbeddingIndex,
VectorDBWithIndex,
)
@ -128,11 +129,12 @@ class FaissIndex(EmbeddingIndex):
# Save updated index
await self._save_index()
async def delete_chunk(self, chunk_id: str) -> None:
if chunk_id not in self.chunk_ids:
async def delete_chunks(self, chunks_for_deletion: list[ChunkForDeletion]) -> None:
chunk_ids = [c.chunk_id for c in chunks_for_deletion]
if not set(chunk_ids).issubset(self.chunk_ids):
return
async with self.chunk_id_lock:
def remove_chunk(chunk_id: str):
index = self.chunk_ids.index(chunk_id)
self.index.remove_ids(np.array([index]))
@ -146,6 +148,10 @@ class FaissIndex(EmbeddingIndex):
self.chunk_by_index = new_chunk_by_index
self.chunk_ids.pop(index)
async with self.chunk_id_lock:
for chunk_id in chunk_ids:
remove_chunk(chunk_id)
await self._save_index()
async def query_vector(
@ -174,7 +180,9 @@ class FaissIndex(EmbeddingIndex):
k: int,
score_threshold: float,
) -> QueryChunksResponse:
raise NotImplementedError("Keyword search is not supported in FAISS")
raise NotImplementedError(
"Keyword search is not supported - underlying DB FAISS does not support this search mode"
)
async def query_hybrid(
self,
@ -185,7 +193,9 @@ class FaissIndex(EmbeddingIndex):
reranker_type: str,
reranker_params: dict[str, Any] | None = None,
) -> QueryChunksResponse:
raise NotImplementedError("Hybrid search is not supported in FAISS")
raise NotImplementedError(
"Hybrid search is not supported - underlying DB FAISS does not support this search mode"
)
class FaissVectorIOAdapter(OpenAIVectorStoreMixin, VectorIO, VectorDBsProtocolPrivate):
@ -293,8 +303,7 @@ class FaissVectorIOAdapter(OpenAIVectorStoreMixin, VectorIO, VectorDBsProtocolPr
return await index.query_chunks(query, params)
async def delete_chunks(self, store_id: str, chunk_ids: list[str]) -> None:
"""Delete a chunk from a faiss index"""
async def delete_chunks(self, store_id: str, chunks_for_deletion: list[ChunkForDeletion]) -> None:
"""Delete chunks from a faiss index"""
faiss_index = self.cache[store_id].index
for chunk_id in chunk_ids:
await faiss_index.delete_chunk(chunk_id)
await faiss_index.delete_chunks(chunks_for_deletion)

25
llama_stack/providers/inline/vector_io/sqlite_vec/sqlite_vec.py
View file

@ -31,6 +31,7 @@ from llama_stack.providers.utils.memory.openai_vector_store_mixin import OpenAIV
from llama_stack.providers.utils.memory.vector_store import (
RERANKER_TYPE_RRF,
RERANKER_TYPE_WEIGHTED,
ChunkForDeletion,
EmbeddingIndex,
VectorDBWithIndex,
)
@ -426,34 +427,36 @@ class SQLiteVecIndex(EmbeddingIndex):
return QueryChunksResponse(chunks=chunks, scores=scores)
async def delete_chunk(self, chunk_id: str) -> None:
async def delete_chunks(self, chunks_for_deletion: list[ChunkForDeletion]) -> None:
"""Remove a chunk from the SQLite vector store."""
chunk_ids = [c.chunk_id for c in chunks_for_deletion]
def _delete_chunk():
def _delete_chunks():
connection = _create_sqlite_connection(self.db_path)
cur = connection.cursor()
try:
cur.execute("BEGIN TRANSACTION")
# Delete from metadata table
cur.execute(f"DELETE FROM {self.metadata_table} WHERE id = ?", (chunk_id,))
placeholders = ",".join("?" * len(chunk_ids))
cur.execute(f"DELETE FROM {self.metadata_table} WHERE id IN ({placeholders})", chunk_ids)
# Delete from vector table
cur.execute(f"DELETE FROM {self.vector_table} WHERE id = ?", (chunk_id,))
cur.execute(f"DELETE FROM {self.vector_table} WHERE id IN ({placeholders})", chunk_ids)
# Delete from FTS table
cur.execute(f"DELETE FROM {self.fts_table} WHERE id = ?", (chunk_id,))
cur.execute(f"DELETE FROM {self.fts_table} WHERE id IN ({placeholders})", chunk_ids)
connection.commit()
except Exception as e:
connection.rollback()
logger.error(f"Error deleting chunk {chunk_id}: {e}")
logger.error(f"Error deleting chunks: {e}")
raise
finally:
cur.close()
connection.close()
await asyncio.to_thread(_delete_chunk)
await asyncio.to_thread(_delete_chunks)
class SQLiteVecVectorIOAdapter(OpenAIVectorStoreMixin, VectorIO, VectorDBsProtocolPrivate):
@ -551,12 +554,10 @@ class SQLiteVecVectorIOAdapter(OpenAIVectorStoreMixin, VectorIO, VectorDBsProtoc
raise VectorStoreNotFoundError(vector_db_id)
return await index.query_chunks(query, params)
async def delete_chunks(self, store_id: str, chunk_ids: list[str]) -> None:
"""Delete a chunk from a sqlite_vec index."""
async def delete_chunks(self, store_id: str, chunks_for_deletion: list[ChunkForDeletion]) -> None:
"""Delete chunks from a sqlite_vec index."""
index = await self._get_and_cache_vector_db_index(store_id)
if not index:
raise VectorStoreNotFoundError(store_id)
for chunk_id in chunk_ids:
# Use the index's delete_chunk method
await index.index.delete_chunk(chunk_id)
await index.index.delete_chunks(chunks_for_deletion)