impls -> inline, adapters -> remote (#381)

llama_stack/providers/remote/safety/__init__.py

@@ -0,0 +1,5 @@
+# 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.

llama_stack/providers/remote/safety/bedrock/__init__.py

@@ -0,0 +1,18 @@
+# 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.
+
+
+from typing import Any
+
+from .config import BedrockSafetyConfig
+
+
+async def get_adapter_impl(config: BedrockSafetyConfig, _deps) -> Any:
+    from .bedrock import BedrockSafetyAdapter
+
+    impl = BedrockSafetyAdapter(config)
+    await impl.initialize()
+    return impl

llama_stack/providers/remote/safety/bedrock/bedrock.py

@@ -0,0 +1,119 @@
+# 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
+import logging
+
+from typing import Any, Dict, List
+
+from llama_stack.apis.safety import *  # noqa
+from llama_models.llama3.api.datatypes import *  # noqa: F403
+from llama_stack.providers.datatypes import ShieldsProtocolPrivate
+from llama_stack.providers.utils.bedrock.client import create_bedrock_client
+
+from .config import BedrockSafetyConfig
+
+
+logger = logging.getLogger(__name__)
+
+
+BEDROCK_SUPPORTED_SHIELDS = [
+    ShieldType.generic_content_shield.value,
+]
+
+
+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: ShieldDef) -> None:
+        raise ValueError("Registering dynamic shields is not supported")
+
+    async def list_shields(self) -> List[ShieldDef]:
+        response = self.bedrock_client.list_guardrails()
+        shields = []
+        for guardrail in response["guardrails"]:
+            # populate the shield def with the guardrail id and version
+            shield_def = ShieldDef(
+                identifier=guardrail["id"],
+                shield_type=ShieldType.generic_content_shield.value,
+                params={
+                    "guardrailIdentifier": guardrail["id"],
+                    "guardrailVersion": guardrail["version"],
+                },
+            )
+            self.registered_shields.append(shield_def)
+            shields.append(shield_def)
+        return shields
+
+    async def run_shield(
+        self, identifier: str, messages: List[Message], params: Dict[str, Any] = None
+    ) -> RunShieldResponse:
+        shield_def = await self.shield_store.get_shield(identifier)
+        if not shield_def:
+            raise ValueError(f"Unknown shield {identifier}")
+
+        """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?"
+                }
+            }
+        ]```
+        However the incoming messages are of this type UserMessage(content=....) coming from
+        https://github.com/meta-llama/llama-models/blob/main/models/llama3/api/datatypes.py
+
+        They contain content, role . For now we will extract the content and default the "qualifiers": ["query"]
+        """
+
+        shield_params = shield_def.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_params["guardrailIdentifier"],
+            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()

llama_stack/providers/remote/safety/bedrock/config.py

@@ -0,0 +1,15 @@
+# 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.
+
+
+from llama_models.schema_utils import json_schema_type
+
+from llama_stack.providers.utils.bedrock.config import BedrockBaseConfig
+
+
+@json_schema_type
+class BedrockSafetyConfig(BedrockBaseConfig):
+    pass

llama_stack/providers/remote/safety/sample/__init__.py

@@ -0,0 +1,17 @@
+# 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.
+
+from typing import Any
+
+from .config import SampleConfig
+
+
+async def get_adapter_impl(config: SampleConfig, _deps) -> Any:
+    from .sample import SampleSafetyImpl
+
+    impl = SampleSafetyImpl(config)
+    await impl.initialize()
+    return impl

llama_stack/providers/remote/safety/sample/config.py

@@ -0,0 +1,12 @@
+# 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.
+
+from pydantic import BaseModel
+
+
+class SampleConfig(BaseModel):
+    host: str = "localhost"
+    port: int = 9999

llama_stack/providers/remote/safety/sample/sample.py

@@ -0,0 +1,23 @@
+# 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.
+
+from .config import SampleConfig
+
+
+from llama_stack.apis.safety import *  # noqa: F403
+
+
+class SampleSafetyImpl(Safety):
+    def __init__(self, config: SampleConfig):
+        self.config = config
+
+    async def register_shield(self, shield: ShieldDef) -> None:
+        # these are the safety shields the Llama Stack will use to route requests to this provider
+        # perform validation here if necessary
+        pass
+
+    async def initialize(self):
+        pass

llama_stack/providers/remote/safety/together/__init__.py

@@ -0,0 +1,18 @@
+# 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.
+
+from .config import TogetherProviderDataValidator, TogetherSafetyConfig  # noqa: F401
+
+
+async def get_adapter_impl(config: TogetherSafetyConfig, _deps):
+    from .together import TogetherSafetyImpl
+
+    assert isinstance(
+        config, TogetherSafetyConfig
+    ), f"Unexpected config type: {type(config)}"
+    impl = TogetherSafetyImpl(config)
+    await impl.initialize()
+    return impl

llama_stack/providers/remote/safety/together/config.py

@@ -0,0 +1,26 @@
+# 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.
+
+from typing import Optional
+
+from llama_models.schema_utils import json_schema_type
+from pydantic import BaseModel, Field
+
+
+class TogetherProviderDataValidator(BaseModel):
+    together_api_key: str
+
+
+@json_schema_type
+class TogetherSafetyConfig(BaseModel):
+    url: str = Field(
+        default="https://api.together.xyz/v1",
+        description="The URL for the Together AI server",
+    )
+    api_key: Optional[str] = Field(
+        default=None,
+        description="The Together AI API Key (default for the distribution, if any)",
+    )

llama_stack/providers/remote/safety/together/together.py

@@ -0,0 +1,101 @@
+# 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.
+from together import Together
+
+from llama_models.llama3.api.datatypes import *  # noqa: F403
+from llama_stack.apis.safety import *  # noqa: F403
+from llama_stack.distribution.request_headers import NeedsRequestProviderData
+from llama_stack.providers.datatypes import ShieldsProtocolPrivate
+
+from .config import TogetherSafetyConfig
+
+
+TOGETHER_SHIELD_MODEL_MAP = {
+    "llama_guard": "meta-llama/Meta-Llama-Guard-3-8B",
+    "Llama-Guard-3-8B": "meta-llama/Meta-Llama-Guard-3-8B",
+    "Llama-Guard-3-11B-Vision": "meta-llama/Llama-Guard-3-11B-Vision-Turbo",
+}
+
+
+class TogetherSafetyImpl(Safety, NeedsRequestProviderData, ShieldsProtocolPrivate):
+    def __init__(self, config: TogetherSafetyConfig) -> None:
+        self.config = config
+
+    async def initialize(self) -> None:
+        pass
+
+    async def shutdown(self) -> None:
+        pass
+
+    async def register_shield(self, shield: ShieldDef) -> None:
+        raise ValueError("Registering dynamic shields is not supported")
+
+    async def list_shields(self) -> List[ShieldDef]:
+        return [
+            ShieldDef(
+                identifier=ShieldType.llama_guard.value,
+                shield_type=ShieldType.llama_guard.value,
+                params={},
+            )
+        ]
+
+    async def run_shield(
+        self, identifier: str, messages: List[Message], params: Dict[str, Any] = None
+    ) -> RunShieldResponse:
+        shield_def = await self.shield_store.get_shield(identifier)
+        if not shield_def:
+            raise ValueError(f"Unknown shield {identifier}")
+
+        model = shield_def.params.get("model", "llama_guard")
+        if model not in TOGETHER_SHIELD_MODEL_MAP:
+            raise ValueError(f"Unsupported safety model: {model}")
+
+        together_api_key = None
+        if self.config.api_key is not None:
+            together_api_key = self.config.api_key
+        else:
+            provider_data = self.get_request_provider_data()
+            if provider_data is None or not provider_data.together_api_key:
+                raise ValueError(
+                    'Pass Together API Key in the header X-LlamaStack-ProviderData as { "together_api_key": <your api key>}'
+                )
+            together_api_key = provider_data.together_api_key
+
+        # messages can have role assistant or user
+        api_messages = []
+        for message in messages:
+            if message.role in (Role.user.value, Role.assistant.value):
+                api_messages.append({"role": message.role, "content": message.content})
+
+        violation = await get_safety_response(
+            together_api_key, TOGETHER_SHIELD_MODEL_MAP[model], api_messages
+        )
+        return RunShieldResponse(violation=violation)
+
+
+async def get_safety_response(
+    api_key: str, model_name: str, messages: List[Dict[str, str]]
+) -> Optional[SafetyViolation]:
+    client = Together(api_key=api_key)
+    response = client.chat.completions.create(messages=messages, model=model_name)
+    if len(response.choices) == 0:
+        return None
+
+    response_text = response.choices[0].message.content
+    if response_text == "safe":
+        return None
+
+    parts = response_text.split("\n")
+    if len(parts) != 2:
+        return None
+
+    if parts[0] == "unsafe":
+        return SafetyViolation(
+            violation_level=ViolationLevel.ERROR,
+            metadata={"violation_type": parts[1]},
+        )
+
+    return None