125 lines
4.2 KiB
Python
125 lines
4.2 KiB
Python
import json
|
||
import sys
|
||
from typing import List, Dict
|
||
|
||
from fastchat.conversation import Conversation
|
||
from fastchat import conversation as conv
|
||
from server.model_workers.base import *
|
||
from server.model_workers.base import ApiEmbeddingsParams
|
||
from configs import logger, log_verbose
|
||
|
||
|
||
class OpenAIWorker(ApiModelWorker):
|
||
"""
|
||
支持 OpenAI 格式 API 的 Worker，用于 embedding 和 chat
|
||
"""
|
||
DEFAULT_EMBED_MODEL = "text-embedding-ada-002"
|
||
|
||
def __init__(
|
||
self,
|
||
*,
|
||
model_names: List[str] = ["openai-api"],
|
||
controller_addr: str = None,
|
||
worker_addr: str = None,
|
||
**kwargs,
|
||
):
|
||
kwargs.update(model_names=model_names, controller_addr=controller_addr, worker_addr=worker_addr)
|
||
kwargs.setdefault("context_len", 8192)
|
||
super().__init__(**kwargs)
|
||
|
||
def do_chat(self, params: ApiChatParams) -> Dict:
|
||
from openai import OpenAI
|
||
params.load_config(self.model_names[0])
|
||
|
||
client = OpenAI(
|
||
api_key=params.api_key,
|
||
base_url=params.api_base_url,
|
||
)
|
||
|
||
if log_verbose:
|
||
logger.info(f'{self.__class__.__name__}:params: {params}')
|
||
|
||
try:
|
||
response = client.chat.completions.create(
|
||
model=params.version or self.model_names[0],
|
||
messages=params.messages,
|
||
temperature=params.temperature,
|
||
max_tokens=params.max_tokens,
|
||
top_p=params.top_p,
|
||
stream=True,
|
||
)
|
||
for chunk in response:
|
||
if chunk.choices:
|
||
delta = chunk.choices[0].delta
|
||
if delta.content:
|
||
yield {
|
||
"error_code": 0,
|
||
"text": delta.content,
|
||
}
|
||
except Exception as e:
|
||
logger.error(f"OpenAI API 请求错误: {e}")
|
||
yield {
|
||
"error_code": 500,
|
||
"text": f"请求错误: {str(e)}",
|
||
}
|
||
|
||
def do_embeddings(self, params: ApiEmbeddingsParams) -> Dict:
|
||
from openai import OpenAI
|
||
# embed_texts 里用 worker_class() 默认 model_names 为 ["openai-api"]，会错加载成 openai-api 的 base_url；
|
||
# 在线嵌入应使用 ONLINE_LLM_MODEL 的键（如 bge-m3-api），由调用方写入 params.worker_name。
|
||
params.load_config(params.worker_name or self.model_names[0])
|
||
|
||
client = OpenAI(
|
||
api_key=params.api_key,
|
||
base_url=params.api_base_url,
|
||
)
|
||
|
||
if log_verbose:
|
||
logger.info(f'{self.__class__.__name__}:params: {params}')
|
||
|
||
try:
|
||
# OpenAI embedding API 每次最多处理 2048 个文本，这里分批处理
|
||
result = []
|
||
batch_size = 100
|
||
for i in range(0, len(params.texts), batch_size):
|
||
batch_texts = params.texts[i:i+batch_size]
|
||
response = client.embeddings.create(
|
||
model=params.embed_model or self.DEFAULT_EMBED_MODEL,
|
||
input=batch_texts,
|
||
encoding_format="float",
|
||
)
|
||
embeddings = [item.embedding for item in response.data]
|
||
result.extend(embeddings)
|
||
|
||
return {"code": 200, "data": result}
|
||
except Exception as e:
|
||
logger.error(f"OpenAI Embedding API 请求错误: {e}")
|
||
return {
|
||
"code": 500,
|
||
"msg": f"Embedding 请求错误: {str(e)}",
|
||
}
|
||
|
||
def make_conv_template(self, conv_template: str = None, model_path: str = None) -> Conversation:
|
||
return conv.Conversation(
|
||
name=self.model_names[0],
|
||
system_message="You are a helpful assistant.",
|
||
messages=[],
|
||
roles=["user", "assistant", "system"],
|
||
sep="\n",
|
||
stop_str="",
|
||
)
|
||
|
||
|
||
if __name__ == "__main__":
|
||
import uvicorn
|
||
from server.utils import MakeFastAPIOffline
|
||
from fastchat.serve.model_worker import app
|
||
|
||
worker = OpenAIWorker(
|
||
controller_addr="http://127.0.0.1:20001",
|
||
worker_addr="http://127.0.0.1:20008",
|
||
)
|
||
sys.modules["fastchat.serve.model_worker"].worker = worker
|
||
MakeFastAPIOffline(app)
|
||
uvicorn.run(app, port=20008)
|