llama-stack-mirror/llama_stack/providers/remote/safety/bedrock/bedrock.py
Francisco Javier Arceo 6620b625f1 adding logo and favicon
Signed-off-by: Francisco Javier Arceo <farceo@redhat.com>

chore: Enable keyword search for Milvus inline (#3073)

With https://github.com/milvus-io/milvus-lite/pull/294 - Milvus Lite
supports keyword search using BM25. While introducing keyword search we
had explicitly disabled it for inline milvus. This PR removes the need
for the check, and enables `inline::milvus` for tests.

<!-- If resolving an issue, uncomment and update the line below -->
<!-- Closes #[issue-number] -->

Run llama stack with `inline::milvus` enabled:

```
pytest tests/integration/vector_io/test_openai_vector_stores.py::test_openai_vector_store_search_modes --stack-config=http://localhost:8321 --embedding-model=all-MiniLM-L6-v2 -v
```

```
INFO     2025-08-07 17:06:20,932 tests.integration.conftest:64 tests: Setting DISABLE_CODE_SANDBOX=1 for macOS
=========================================================================================== test session starts ============================================================================================
platform darwin -- Python 3.12.11, pytest-7.4.4, pluggy-1.5.0 -- /Users/vnarsing/miniconda3/envs/stack-client/bin/python
cachedir: .pytest_cache
metadata: {'Python': '3.12.11', 'Platform': 'macOS-14.7.6-arm64-arm-64bit', 'Packages': {'pytest': '7.4.4', 'pluggy': '1.5.0'}, 'Plugins': {'asyncio': '0.23.8', 'cov': '6.0.0', 'timeout': '2.2.0', 'socket': '0.7.0', 'html': '3.1.1', 'langsmith': '0.3.39', 'anyio': '4.8.0', 'metadata': '3.0.0'}}
rootdir: /Users/vnarsing/go/src/github/meta-llama/llama-stack
configfile: pyproject.toml
plugins: asyncio-0.23.8, cov-6.0.0, timeout-2.2.0, socket-0.7.0, html-3.1.1, langsmith-0.3.39, anyio-4.8.0, metadata-3.0.0
asyncio: mode=Mode.AUTO
collected 3 items

tests/integration/vector_io/test_openai_vector_stores.py::test_openai_vector_store_search_modes[None-None-all-MiniLM-L6-v2-None-384-vector] PASSED                                                   [ 33%]
tests/integration/vector_io/test_openai_vector_stores.py::test_openai_vector_store_search_modes[None-None-all-MiniLM-L6-v2-None-384-keyword] PASSED                                                  [ 66%]
tests/integration/vector_io/test_openai_vector_stores.py::test_openai_vector_store_search_modes[None-None-all-MiniLM-L6-v2-None-384-hybrid] PASSED                                                   [100%]

============================================================================================ 3 passed in 4.75s =============================================================================================
```

Signed-off-by: Varsha Prasad Narsing <varshaprasad96@gmail.com>
Co-authored-by: Francisco Arceo <arceofrancisco@gmail.com>

chore: Fixup main pre commit (#3204)

build: Bump version to 0.2.18

chore: Faster npm pre-commit (#3206)

Adds npm to pre-commit.yml installation and caches ui
Removes node installation during pre-commit.

<!-- If resolving an issue, uncomment and update the line below -->
<!-- Closes #[issue-number] -->

<!-- Describe the tests you ran to verify your changes with result
summaries. *Provide clear instructions so the plan can be easily
re-executed.* -->

Signed-off-by: Francisco Javier Arceo <farceo@redhat.com>

chiecking in for tonight, wip moving to agents api

Signed-off-by: Francisco Javier Arceo <farceo@redhat.com>

remove log

Signed-off-by: Francisco Javier Arceo <farceo@redhat.com>

updated

Signed-off-by: Francisco Javier Arceo <farceo@redhat.com>

fix: disable ui-prettier & ui-eslint (#3207)

chore(pre-commit): add pre-commit hook to enforce llama_stack logger usage (#3061)

This PR adds a step in pre-commit to enforce using `llama_stack` logger.

Currently, various parts of the code base uses different loggers. As a
custom `llama_stack` logger exist and used in the codebase, it is better
to standardize its utilization.

Signed-off-by: Mustafa Elbehery <melbeher@redhat.com>
Co-authored-by: Matthew Farrellee <matt@cs.wisc.edu>

fix: fix ```openai_embeddings``` for asymmetric embedding NIMs (#3205)

NVIDIA asymmetric embedding models (e.g.,
`nvidia/llama-3.2-nv-embedqa-1b-v2`) require an `input_type` parameter
not present in the standard OpenAI embeddings API. This PR adds the
`input_type="query"` as default and updates the documentation to suggest
using the `embedding` API for passage embeddings.

<!-- If resolving an issue, uncomment and update the line below -->
Resolves #2892

```
pytest -s -v tests/integration/inference/test_openai_embeddings.py   --stack-config="inference=nvidia"   --embedding-model="nvidia/llama-3.2-nv-embedqa-1b-v2"   --env NVIDIA_API_KEY={nvidia_api_key}   --env NVIDIA_BASE_URL="https://integrate.api.nvidia.com"
```

cleaning up

Signed-off-by: Francisco Javier Arceo <farceo@redhat.com>

updating session manager to cache messages locally

Signed-off-by: Francisco Javier Arceo <farceo@redhat.com>

fix linter

Signed-off-by: Francisco Javier Arceo <farceo@redhat.com>

more cleanup

Signed-off-by: Francisco Javier Arceo <farceo@redhat.com>
2025-08-21 16:06:30 -04:00

111 lines
4.1 KiB
Python

# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
# This source code is licensed under the terms described in the LICENSE file in
# the root directory of this source tree.
import json
from typing import Any
from llama_stack.apis.inference import Message
from llama_stack.apis.safety import (
RunShieldResponse,
Safety,
SafetyViolation,
ViolationLevel,
)
from llama_stack.apis.shields import Shield
from llama_stack.log import get_logger
from llama_stack.providers.datatypes import ShieldsProtocolPrivate
from llama_stack.providers.utils.bedrock.client import create_bedrock_client
from .config import BedrockSafetyConfig
logger = get_logger(name=__name__, category="safety")
class BedrockSafetyAdapter(Safety, ShieldsProtocolPrivate):
def __init__(self, config: BedrockSafetyConfig) -> None:
self.config = config
self.registered_shields = []
async def initialize(self) -> None:
try:
self.bedrock_runtime_client = create_bedrock_client(self.config)
self.bedrock_client = create_bedrock_client(self.config, "bedrock")
except Exception as e:
raise RuntimeError("Error initializing BedrockSafetyAdapter") from e
async def shutdown(self) -> None:
pass
async def register_shield(self, shield: Shield) -> None:
response = self.bedrock_client.list_guardrails(
guardrailIdentifier=shield.provider_resource_id,
)
if (
not response["guardrails"]
or len(response["guardrails"]) == 0
or response["guardrails"][0]["version"] != shield.params["guardrailVersion"]
):
raise ValueError(
f"Shield {shield.provider_resource_id} with version {shield.params['guardrailVersion']} not found in Bedrock"
)
async def unregister_shield(self, identifier: str) -> None:
pass
async def run_shield(
self, shield_id: str, messages: list[Message], params: dict[str, Any] = None
) -> RunShieldResponse:
shield = await self.shield_store.get_shield(shield_id)
if not shield:
raise ValueError(f"Shield {shield_id} not found")
"""
This is the implementation for the bedrock guardrails. The input to the guardrails is to be of this format
```content = [
{
"text": {
"text": "Is the AB503 Product a better investment than the S&P 500?"
}
}
]```
Incoming messages contain content, role . For now we will extract the content and
default the "qualifiers": ["query"]
"""
shield_params = shield.params
logger.debug(f"run_shield::{shield_params}::messages={messages}")
# - convert the messages into format Bedrock expects
content_messages = []
for message in messages:
content_messages.append({"text": {"text": message.content}})
logger.debug(f"run_shield::final:messages::{json.dumps(content_messages, indent=2)}:")
response = self.bedrock_runtime_client.apply_guardrail(
guardrailIdentifier=shield.provider_resource_id,
guardrailVersion=shield_params["guardrailVersion"],
source="OUTPUT", # or 'INPUT' depending on your use case
content=content_messages,
)
if response["action"] == "GUARDRAIL_INTERVENED":
user_message = ""
metadata = {}
for output in response["outputs"]:
# guardrails returns a list - however for this implementation we will leverage the last values
user_message = output["text"]
for assessment in response["assessments"]:
# guardrails returns a list - however for this implementation we will leverage the last values
metadata = dict(assessment)
return RunShieldResponse(
violation=SafetyViolation(
user_message=user_message,
violation_level=ViolationLevel.ERROR,
metadata=metadata,
)
)
return RunShieldResponse()