phoenix-oss/llama-stack-mirror
0
0
Fork 1
mirror of https://github.com/meta-llama/llama-stack.git synced 2025-07-02 20:40:36 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Remove the "ShieldType" concept (#430)

Browse source 
# What does this PR do?

This PR kills the notion of "ShieldType". The impetus for this is the
realization:

> Why is keyword llama-guard appearing so many times everywhere,
sometimes with hyphens, sometimes with underscores?

Now that we have a notion of "provider specific resource identifiers"
and "user specific aliases" for those and the fact that this works with
models ("Llama3.1-8B-Instruct" <> "fireworks/llama-3pv1-..."), we can
follow the same rules for Shields.

So each Safety provider can make up a notion of identifiers it has
registered. This already happens with Bedrock correctly. We just
generalize it for Llama Guard, Prompt Guard, etc.

For Llama Guard, we further simplify by just adopting the underlying
model name itself as the identifier! No confusion necessary.

While doing this, I noticed a bug in our DistributionRegistry where we
weren't scoping identifiers by type. Fixed.

## Feature/Issue validation/testing/test plan

Ran (inference, safety, memory, agents) tests with ollama and fireworks
providers.
This commit is contained in:
Ashwin Bharambe 2024-11-12 12:37:24 -08:00 • committed by GitHub
parent 09269e2a44
commit 983d6ce2df
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
26 changed files with 150 additions and 209 deletions
Download patch file Download diff file Expand all files Collapse all files

24
llama_stack/providers/tests/safety/fixtures.py
View file

@ -9,7 +9,7 @@ import pytest_asyncio
from llama_stack.apis.models import ModelInput
from llama_stack.apis.shields import ShieldInput, ShieldType
from llama_stack.apis.shields import ShieldInput
from llama_stack.distribution.datatypes import Api, Provider
from llama_stack.providers.inline.safety.llama_guard import LlamaGuardConfig
@ -41,7 +41,7 @@ def safety_llama_guard(safety_model) -> ProviderFixture:
Provider(
provider_id="inline::llama-guard",
provider_type="inline::llama-guard",
config=LlamaGuardConfig(model=safety_model).model_dump(),
config=LlamaGuardConfig().model_dump(),
)
],
)
@ -101,6 +101,8 @@ async def safety_stack(inference_model, safety_model, request):
shield_provider_type = safety_fixture.providers[0].provider_type
shield_input = get_shield_to_register(shield_provider_type, safety_model)
print(f"inference_model: {inference_model}")
print(f"shield_input = {shield_input}")
impls = await resolve_impls_for_test_v2(
[Api.safety, Api.shields, Api.inference],
providers,
@ -114,20 +116,14 @@ async def safety_stack(inference_model, safety_model, request):
def get_shield_to_register(provider_type: str, safety_model: str) -> ShieldInput:
shield_config = {}
shield_type = ShieldType.llama_guard
identifier = "llama_guard"
if provider_type == "meta-reference":
shield_config["model"] = safety_model
elif provider_type == "remote::together":
shield_config["model"] = safety_model
elif provider_type == "remote::bedrock":
if provider_type == "remote::bedrock":
identifier = get_env_or_fail("BEDROCK_GUARDRAIL_IDENTIFIER")
shield_config["guardrailVersion"] = get_env_or_fail("BEDROCK_GUARDRAIL_VERSION")
shield_type = ShieldType.generic_content_shield
params = {"guardrailVersion": get_env_or_fail("BEDROCK_GUARDRAIL_VERSION")}
else:
params = {}
identifier = safety_model
return ShieldInput(
shield_id=identifier,
shield_type=shield_type,
params=shield_config,
params=params,
)