phoenix-oss/llama-stack
1
0
Fork 0
forked from phoenix-oss/llama-stack-mirror
Code Pull requests Releases Packages 2 Activity Actions

Remove "routing_table" and "routing_key" concepts for the user (#201)

Browse source 
This PR makes several core changes to the developer experience surrounding Llama Stack.

Background: PR #92 introduced the notion of "routing" to the Llama Stack. It introduces three object types: (1) models, (2) shields and (3) memory banks. Each of these objects can be associated with a distinct provider. So you can get model A to be inferenced locally while model B, C can be inference remotely (e.g.)

However, this had a few drawbacks:

you could not address the provider instances -- i.e., if you configured "meta-reference" with a given model, you could not assign an identifier to this instance which you could re-use later.
the above meant that you could not register a "routing_key" (e.g. model) dynamically and say "please use this existing provider I have already configured" for a new model.
the terms "routing_table" and "routing_key" were exposed directly to the user. in my view, this is way too much overhead for a new user (which almost everyone is.) people come to the stack wanting to do ML and encounter a completely unexpected term.
What this PR does: This PR structures the run config with only a single prominent key:

- providers
Providers are instances of configured provider types. Here's an example which shows two instances of the remote::tgi provider which are serving two different models.

providers:
  inference:
  - provider_id: foo
    provider_type: remote::tgi
    config: { ... }
  - provider_id: bar
    provider_type: remote::tgi
    config: { ... }
Secondly, the PR adds dynamic registration of { models | shields | memory_banks } to the API surface. The distribution still acts like a "routing table" (as previously) except that it asks the backing providers for a listing of these objects. For example it asks a TGI or Ollama inference adapter what models it is serving. Only the models that are being actually served can be requested by the user for inference. Otherwise, the Stack server will throw an error.

When dynamically registering these objects, you can use the provider IDs shown above. Info about providers can be obtained using the Api.inspect set of endpoints (/providers, /routes, etc.)

The above examples shows the correspondence between inference providers and models registry items. Things work similarly for the safety <=> shields and memory <=> memory_banks pairs.

Registry: This PR also makes it so that Providers need to implement additional methods for registering and listing objects. For example, each Inference provider is now expected to implement the ModelsProtocolPrivate protocol (naming is not great!) which consists of two methods

register_model
list_models
The goal is to inform the provider that a certain model needs to be supported so the provider can make any relevant backend changes if needed (or throw an error if the model cannot be supported.)

There are many other cleanups included some of which are detailed in a follow-up comment.
This commit is contained in:
Ashwin Bharambe 2024-10-10 10:24:13 -07:00 • committed by GitHub
parent 8c3010553f
commit 6bb57e72a7
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
93 changed files with 4697 additions and 4457 deletions
Download patch file Download diff file Expand all files Collapse all files

99
llama_stack/providers/adapters/safety/bedrock/bedrock.py
View file

@ -7,14 +7,13 @@
import json
import logging
import traceback
from typing import Any, Dict, List
import boto3
from llama_stack.apis.safety import * # noqa
from llama_models.llama3.api.datatypes import * # noqa: F403
from llama_stack.distribution.datatypes import RoutableProvider
from llama_stack.providers.datatypes import ShieldsProtocolPrivate
from .config import BedrockSafetyConfig
@ -22,16 +21,17 @@ from .config import BedrockSafetyConfig
logger = logging.getLogger(__name__)
SUPPORTED_SHIELD_TYPES = [
"bedrock_guardrail",
BEDROCK_SUPPORTED_SHIELDS = [
ShieldType.generic_content_shield.value,
]
class BedrockSafetyAdapter(Safety, RoutableProvider):
class BedrockSafetyAdapter(Safety, ShieldsProtocolPrivate):
def __init__(self, config: BedrockSafetyConfig) -> None:
if not config.aws_profile:
raise ValueError(f"Missing boto_client aws_profile in model info::{config}")
self.config = config
self.registered_shields = []
async def initialize(self) -> None:
try:
@ -45,16 +45,23 @@ class BedrockSafetyAdapter(Safety, RoutableProvider):
async def shutdown(self) -> None:
pass
async def validate_routing_keys(self, routing_keys: List[str]) -> None:
for key in routing_keys:
if key not in SUPPORTED_SHIELD_TYPES:
raise ValueError(f"Unknown safety shield type: {key}")
async def register_shield(self, shield: ShieldDef) -> None:
raise ValueError("Registering dynamic shields is not supported")
async def list_shields(self) -> List[ShieldDef]:
raise NotImplementedError(
"""
`list_shields` not implemented; this should read all guardrails from
bedrock and populate guardrailId and guardrailVersion in the ShieldDef.
"""
)
async def run_shield(
self, shield_type: str, messages: List[Message], params: Dict[str, Any] = None
) -> RunShieldResponse:
if shield_type not in SUPPORTED_SHIELD_TYPES:
raise ValueError(f"Unknown safety shield type: {shield_type}")
shield_def = await self.shield_store.get_shield(shield_type)
if not shield_def:
raise ValueError(f"Unknown shield {shield_type}")
"""This is the implementation for the bedrock guardrails. The input to the guardrails is to be of this format
```content = [
@ -69,52 +76,38 @@ class BedrockSafetyAdapter(Safety, RoutableProvider):
They contain content, role . For now we will extract the content and default the "qualifiers": ["query"]
"""
try:
logger.debug(f"run_shield::{params}::messages={messages}")
if "guardrailIdentifier" not in params:
raise RuntimeError(
"Error running request for BedrockGaurdrails:Missing GuardrailID in request"
)
if "guardrailVersion" not in params:
raise RuntimeError(
"Error running request for BedrockGaurdrails:Missing guardrailVersion in request"
)
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)}:"
)
# - 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.boto_client.apply_guardrail(
guardrailIdentifier=params.get("guardrailIdentifier"),
guardrailVersion=params.get("guardrailVersion"),
source="OUTPUT", # or 'INPUT' depending on your use case
content=content_messages,
)
logger.debug(f"run_shield:: response: {response}::")
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 SafetyViolation(
user_message=user_message,
violation_level=ViolationLevel.ERROR,
metadata=metadata,
)
response = self.boto_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)
except Exception:
error_str = traceback.format_exc()
logger.error(
f"Error in apply_guardrails:{error_str}:: RETURNING None !!!!!"
return SafetyViolation(
user_message=user_message,
violation_level=ViolationLevel.ERROR,
metadata=metadata,
)
return None