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

chore: remove llama_models.llama3.api imports from providers (#1107)

Browse source 
There should be a choke-point for llama3.api imports -- this is the
prompt adapter. Creating a ChatFormat() object on demand is inexpensive.
The underlying Tokenizer is a singleton anyway.
This commit is contained in:
Ashwin Bharambe 2025-02-19 19:01:29 -08:00 • committed by GitHub
parent e9b8259cf9
commit cdcbeb005b
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
13 changed files with 77 additions and 113 deletions
Download patch file Download diff file Expand all files Collapse all files

15
llama_stack/providers/remote/inference/fireworks/fireworks.py
View file

@ -7,8 +7,6 @@
from typing import AsyncGenerator, List, Optional, Union
from fireworks.client import Fireworks
from llama_models.llama3.api.chat_format import ChatFormat
from llama_models.llama3.api.tokenizer import Tokenizer
from llama_stack.apis.common.content_types import InterleavedContent
from llama_stack.apis.inference import (
@ -56,7 +54,6 @@ class FireworksInferenceAdapter(ModelRegistryHelper, Inference, NeedsRequestProv
def __init__(self, config: FireworksImplConfig) -> None:
ModelRegistryHelper.__init__(self, MODEL_ALIASES)
self.config = config
self.formatter = ChatFormat(Tokenizer.get_instance())
async def initialize(self) -> None:
pass
@ -105,7 +102,7 @@ class FireworksInferenceAdapter(ModelRegistryHelper, Inference, NeedsRequestProv
async def _nonstream_completion(self, request: CompletionRequest) -> CompletionResponse:
params = await self._get_params(request)
r = await self._get_client().completion.acreate(**params)
return process_completion_response(r, self.formatter)
return process_completion_response(r)
async def _stream_completion(self, request: CompletionRequest) -> AsyncGenerator:
params = await self._get_params(request)
@ -117,7 +114,7 @@ class FireworksInferenceAdapter(ModelRegistryHelper, Inference, NeedsRequestProv
yield chunk
stream = _to_async_generator()
async for chunk in process_completion_stream_response(stream, self.formatter):
async for chunk in process_completion_stream_response(stream):
yield chunk
def _build_options(
@ -186,7 +183,7 @@ class FireworksInferenceAdapter(ModelRegistryHelper, Inference, NeedsRequestProv
r = await self._get_client().chat.completions.acreate(**params)
else:
r = await self._get_client().completion.acreate(**params)
return process_chat_completion_response(r, self.formatter, request)
return process_chat_completion_response(r, request)
async def _stream_chat_completion(self, request: ChatCompletionRequest) -> AsyncGenerator:
params = await self._get_params(request)
@ -200,7 +197,7 @@ class FireworksInferenceAdapter(ModelRegistryHelper, Inference, NeedsRequestProv
yield chunk
stream = _to_async_generator()
async for chunk in process_chat_completion_stream_response(stream, self.formatter, request):
async for chunk in process_chat_completion_stream_response(stream, request):
yield chunk
async def _get_params(self, request: Union[ChatCompletionRequest, CompletionRequest]) -> dict:
@ -214,11 +211,11 @@ class FireworksInferenceAdapter(ModelRegistryHelper, Inference, NeedsRequestProv
]
else:
input_dict["prompt"] = await chat_completion_request_to_prompt(
request, self.get_llama_model(request.model), self.formatter
request, self.get_llama_model(request.model)
)
else:
assert not media_present, "Fireworks does not support media for Completion requests"
input_dict["prompt"] = await completion_request_to_prompt(request, self.formatter)
input_dict["prompt"] = await completion_request_to_prompt(request)
# Fireworks always prepends with BOS
if "prompt" in input_dict: