本人精力有限,某些模型更新可能无法及时更新,如果遇到问题请提issue,也欢迎有兴趣的大佬提PR
屏蔽不同大模型API的差异,统一用openai api标准格式使用大模型, 也可以用来做api-key的二次分发管理; 配置化管理不同大模型调用参数,让你在使用大模型的时候只需关注 api-key 和 messages
- 支持多种大模型,当前已支持
- openai
- azure open ai
- claude-api 【api申请在等待列表,暂未测试】
- claude-web (将web端功能封装成openai api)
- 智谱ai
- kimi
- bingchat(copilot)
- 百度文心一言
- 讯飞星火
- gemini
- 通义千问
- ...
- 支持stream方式调用
- 支持open ai的第三方代理服务,比如openai-sb等
- 支持在线更新配置
http://0.0.0.0:8090/
(这个前端页面和交互完全是用gpt写的 哈哈) - 支持负载均衡,一个key可轮训/随机/并行等访问多个模型
- 支持按照model_name进行路由
更新日志
2024-04-03
- 支持通义千问
- 优化异常处理
项目的核心配置依赖model-config.json文件,若是没有model-config.json,默认会使用model-config-default.json启动,这时虽然能启动起来,但是因为api-key等没有配置,无法调用成功。
本地新建一个model-config.json文件,根据下边配置文件示例,进行配置, 然后运行以下命令
docker pull tianminghui/openai-style-api
docker run -d -p 8090:8090 --name openai-style-api\
-e ADMIN-TOKEN=admin \
-v /path/to/your/model-config.json:/app/model-config.json \
tianminghui/openai-style-api
/path/to/your/model-config.json
替换成你自己的本地路径
clone本项目,或者下载项目中的docker-compose.yml
文件,修改其中的./model-config.json
路径, 然后运行以下命令
docker-compose up -d
git clone https://github.com/tian-minghui/openai-style-api.git
拉取项目代码cp model-config-default.json model-config.json
并按需修改配置文件model-config.jsonpip install -r requirements.txt
- 运行
python open-api.py
model-config.json 配置文件简单示例
[
{
"token": "f2b7295fc440db7f",
"type": "azure", // azure openai 模型
"config": {
"api_base": "https://xxxx.openai.azure.com/",
"deployment_id": "gpt-35-turbo",
"api_version": "2023-05-15",
"api_key": "xxxxxx",
"temperature": 0.8
}
}
]
- 整个文件是一个json list,可以配置多个模型,只要token不重复就行
- token 自定义的token,后续在请求的时候拿着这个token来请求
- type 类型,表示以下config中的配置是那个模型的,比如 openai,通义千问
- config, 配置openai的api_base, api_key, model等, 针对不用模型有不同的配置(下边有配置示例,更详细配置可以看代码), 此处的配置优先于客户端请求中的配置,比如"temperature": 0.8, 会覆盖请求中的temperature(这里的想法是可以针对同一个模型,调整不同参数,映射成一个新模型)
curl http://localhost:8090/v1/chat/completions \
-H "Content-Type: application/json" \
-H "Authorization: Bearer f2b7295fc440db7f" \
-d '{
"messages": [
{
"role": "system",
"content": "You are a helpful assistant."
},
{
"role": "user",
"content": "Hello!"
}
]
}'
openai<1.0.0 使用如下方式
import openai
openai.api_key = "f2b7295fc440db7f"
openai.api_base = "http://localhost:8090/v1"
completion = openai.ChatCompletion.create(
model="gpt-3.5-turbo", messages=[{"role": "user", "content": "Hello world"}])
print(completion.choices[0].message.content)
openai>=1.0.0使用以下方式调用
import os
from openai import OpenAI
client = OpenAI(
# This is the default and can be omitted
api_key='kimi-GxqT3BlbkFJj',
base_url = 'http://localhost:8090/v1'
)
chat_completion = client.chat.completions.create(
messages=[
{
"role": "user",
"content": "Say this is a test",
}
],
model="gpt-3.5-turbo",
)
print(chat_completion.choices[0].message.content)
[
{
"token": "f2b7295fc440db7f",
"type": "azure", // azure openai 模型
"config": {
"api_base": "https://xxxx.openai.azure.com/",
"deployment_id": "gpt-35-turbo",
"api_version": "2023-05-15",
"api_key": "xxxxxx",
"temperature": 0.8
}
},
{
"token": "GxqT3BlbkFJj",
"type": "openai", // openai 模型
"config": {
"api_base": "https://api.openai.com/v1/",
"api_key": "sk-xxxxxx",
"model": "gpt-3.5-turbo"
}
},
{
"token": "sb-ede1529390cc",
"type": "proxy", // openai 代理
"config": {
"api_base": "https://api.openai-sb.com/v1/",
"api_key": "sb-xxxxxx",
"model": "gpt-3.5-turbo"
}
},
{
"token": "c115c8f5082",
"type": "claude-web", // claude-web
"config": {
"cookie": "xxxxxx",
"proxies": {
"https": "http://localhost:7890"
},
"conversation_id": "xxxxxx",
"prompt": "The information in [] is the context of the conversation. Please ignore the JSON format of the context during the conversation and answer the user's latest conversation: {newMessage} \n {history}",
"single_conversation": true
}
},
{
"token": "7c7aa4a3549f5",
"type": "zhipu-api", // 智谱API
"config": {
"api_key": "xxxxxx",
"model": "chatglm_lite",
"temperature": 0.8,
"top_p": 0.7
}
},
{
"token": "7c7aa4a3549f11",
"type": "xunfei-spark-api", // 讯飞星火API
"config": {
"app_id": "xxxx",
"api_key": "xxxx",
"api_secret": "xxxxxx",
"api_model_version": "v2.0",
"top_k": 5
}
},
{
"token": "7c7aa4a3549f12",
"type": "router", // 路由 可以包含多个模型进行负载均衡
"config": {
"router_strategy": "round-robin", // 路由策略 round-robin 轮询 random 随机
"token_pool": [ // 路由的token池
"7c7aa4a3549f11",
"7c7aa4a3549f5"
]
}
},
{
"token": "7c7aa4a3549f13",
"type": "model-name-router", //根据req中的modelname进行路由, 可以方便的结合ChatGPT-Next-Web
"config": {
"model-2-token": { // 路由的token池
"spark-api-v2.0":"7c7aa4a3549f11",
"chatglm_lite": "7c7aa4a3549f5",
"router-round-robin": "7c7aa4a3549f12"
}
}
},
{
"token": "gemini-7c7aa4a3549f5",
"type": "gemini", // gemini
"config": {
"api_key": "xxxxx",
"proxies": {
"https": "http://localhost:7890"
}
}
},
{
"token": "bing-7c7aa4a3549f5", // 必应
"type": "bing-sydney",
"config": {
"cookie": "xxxxx",
"style": "balanced"
}
},
{
"token":"qwen-111111xxxx", // 通义千问
"type":"qwen",
"config":{
"api_key":"sk-xxxxxxxx",
"model":"qwen-turbo"
}
},
{
"token": "kimi-GxqT3BlbkFJj1", // kimi
"type": "openai", // kimi api与openai相同,因此使用openai就可以
"config": {
"api_base": "https://api.moonshot.cn/v1/",
"api_key": "sk-xxxxxx",
"model": "moonshot-v1-8k"
}
}
]