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

Inference to use provider resource id to register and validate (#428)

Browse source 
This PR changes the way model id gets translated to the final model name
that gets passed through the provider.
Major changes include:
1) Providers are responsible for registering an object and as part of
the registration returning the object with the correct provider specific
name of the model provider_resource_id
2) To help with the common look ups different names a new ModelLookup
class is created.



Tested all inference providers including together, fireworks, vllm,
ollama, meta reference and bedrock
This commit is contained in:
Dinesh Yeduguru 2024-11-12 20:02:00 -08:00 • committed by GitHub
parent e51107e019
commit fdff24e77a
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
21 changed files with 460 additions and 290 deletions
Download patch file Download diff file Expand all files Collapse all files

76
llama_stack/providers/remote/inference/fireworks/fireworks.py
View file

@ -7,14 +7,17 @@
from typing import AsyncGenerator
from fireworks.client import Fireworks
from llama_models.datatypes import CoreModelId
from llama_models.llama3.api.chat_format import ChatFormat
from llama_models.llama3.api.datatypes import Message
from llama_models.llama3.api.tokenizer import Tokenizer
from llama_stack.apis.inference import * # noqa: F403
from llama_stack.distribution.request_headers import NeedsRequestProviderData
from llama_stack.providers.utils.inference.model_registry import ModelRegistryHelper
from llama_stack.providers.utils.inference.model_registry import (
build_model_alias,
ModelRegistryHelper,
)
from llama_stack.providers.utils.inference.openai_compat import (
get_sampling_options,
process_chat_completion_response,
@ -31,25 +34,52 @@ from llama_stack.providers.utils.inference.prompt_adapter import (
from .config import FireworksImplConfig
FIREWORKS_SUPPORTED_MODELS = {
"Llama3.1-8B-Instruct": "fireworks/llama-v3p1-8b-instruct",
"Llama3.1-70B-Instruct": "fireworks/llama-v3p1-70b-instruct",
"Llama3.1-405B-Instruct": "fireworks/llama-v3p1-405b-instruct",
"Llama3.2-1B-Instruct": "fireworks/llama-v3p2-1b-instruct",
"Llama3.2-3B-Instruct": "fireworks/llama-v3p2-3b-instruct",
"Llama3.2-11B-Vision-Instruct": "fireworks/llama-v3p2-11b-vision-instruct",
"Llama3.2-90B-Vision-Instruct": "fireworks/llama-v3p2-90b-vision-instruct",
"Llama-Guard-3-8B": "fireworks/llama-guard-3-8b",
}
model_aliases = [
build_model_alias(
"fireworks/llama-v3p1-8b-instruct",
CoreModelId.llama3_1_8b_instruct.value,
),
build_model_alias(
"fireworks/llama-v3p1-70b-instruct",
CoreModelId.llama3_1_70b_instruct.value,
),
build_model_alias(
"fireworks/llama-v3p1-405b-instruct",
CoreModelId.llama3_1_405b_instruct.value,
),
build_model_alias(
"fireworks/llama-v3p2-1b-instruct",
CoreModelId.llama3_2_3b_instruct.value,
),
build_model_alias(
"fireworks/llama-v3p2-3b-instruct",
CoreModelId.llama3_2_11b_vision_instruct.value,
),
build_model_alias(
"fireworks/llama-v3p2-11b-vision-instruct",
CoreModelId.llama3_2_11b_vision_instruct.value,
),
build_model_alias(
"fireworks/llama-v3p2-90b-vision-instruct",
CoreModelId.llama3_2_90b_vision_instruct.value,
),
build_model_alias(
"fireworks/llama-guard-3-8b",
CoreModelId.llama_guard_3_8b.value,
),
build_model_alias(
"fireworks/llama-guard-3-11b-vision",
CoreModelId.llama_guard_3_11b_vision.value,
),
]
class FireworksInferenceAdapter(
ModelRegistryHelper, Inference, NeedsRequestProviderData
):
def __init__(self, config: FireworksImplConfig) -> None:
ModelRegistryHelper.__init__(
self, stack_to_provider_models_map=FIREWORKS_SUPPORTED_MODELS
)
ModelRegistryHelper.__init__(self, model_aliases)
self.config = config
self.formatter = ChatFormat(Tokenizer.get_instance())
@ -74,15 +104,16 @@ class FireworksInferenceAdapter(
async def completion(
self,
model: str,
model_id: str,
content: InterleavedTextMedia,
sampling_params: Optional[SamplingParams] = SamplingParams(),
response_format: Optional[ResponseFormat] = None,
stream: Optional[bool] = False,
logprobs: Optional[LogProbConfig] = None,
) -> AsyncGenerator:
model = await self.model_store.get_model(model_id)
request = CompletionRequest(
model=model,
model=model.provider_resource_id,
content=content,
sampling_params=sampling_params,
response_format=response_format,
@ -138,7 +169,7 @@ class FireworksInferenceAdapter(
async def chat_completion(
self,
model: str,
model_id: str,
messages: List[Message],
sampling_params: Optional[SamplingParams] = SamplingParams(),
tools: Optional[List[ToolDefinition]] = None,
@ -148,8 +179,9 @@ class FireworksInferenceAdapter(
stream: Optional[bool] = False,
logprobs: Optional[LogProbConfig] = None,
) -> AsyncGenerator:
model = await self.model_store.get_model(model_id)
request = ChatCompletionRequest(
model=model,
model=model.provider_resource_id,
messages=messages,
sampling_params=sampling_params,
tools=tools or [],
@ -207,7 +239,7 @@ class FireworksInferenceAdapter(
]
else:
input_dict["prompt"] = chat_completion_request_to_prompt(
request, self.formatter
request, self.get_llama_model(request.model), self.formatter
)
else:
assert (
@ -221,7 +253,7 @@ class FireworksInferenceAdapter(
input_dict["prompt"] = input_dict["prompt"][len("<|begin_of_text|>") :]
return {
"model": self.map_to_provider_model(request.model),
"model": request.model,
**input_dict,
"stream": request.stream,
**self._build_options(request.sampling_params, request.response_format),
@ -229,7 +261,7 @@ class FireworksInferenceAdapter(
async def embeddings(
self,
model: str,
model_id: str,
contents: List[InterleavedTextMedia],
) -> EmbeddingsResponse:
raise NotImplementedError()