223 lines
6.5 KiB
Markdown
223 lines
6.5 KiB
Markdown
# 本地私有化/大模型接口接入
|
||
|
||
依托于开源的 LLM 与 Embedding 模型，本项目可实现基于开源模型的离线私有部署。此外，本项目也支持 OpenAI API 的调用。
|
||
|
||
## 📜 目录
|
||
- [ 本地私有化模型接入](#本地私有化模型接入)
|
||
- [ 公开大模型接口接入](#公开大模型接口接入)
|
||
- [ 启动大模型服务](#启动大模型服务)
|
||
|
||
## 本地私有化模型接入
|
||
|
||
<br>模型地址配置示例，model_config.py配置修改
|
||
|
||
```bash
|
||
# 建议：走huggingface接入，尽量使用chat模型，不要使用base，无法获取正确输出
|
||
# 注意：当llm_model_dict和VLLM_MODEL_DICT同时存在时，优先启动VLLM_MODEL_DICT中的模型配置
|
||
|
||
# llm_model_dict 配置接入示例如下
|
||
llm_model_dict = {
|
||
"chatglm-6b": {
|
||
"local_model_path": "THUDM/chatglm-6b",
|
||
"api_base_url": "http://localhost:8888/v1", # "name"修改为fastchat服务中的"api_base_url"
|
||
"api_key": "EMPTY"
|
||
}
|
||
}
|
||
|
||
# VLLM_MODEL_DICT 配置接入示例如下
|
||
VLLM_MODEL_DICT = {
|
||
'chatglm2-6b': "THUDM/chatglm-6b",
|
||
}
|
||
|
||
```
|
||
|
||
<br>模型路径填写示例
|
||
|
||
```bash
|
||
# 1、若把模型放到 ~/codefuse-chatbot/llm_models 路径下
|
||
# 若模型地址如下
|
||
model_dir: ~/codefuse-chatbot/llm_models/THUDM/chatglm-6b
|
||
|
||
# 参考配置如下
|
||
llm_model_dict = {
|
||
"chatglm-6b": {
|
||
"local_model_path": "THUDM/chatglm-6b",
|
||
"api_base_url": "http://localhost:8888/v1", # "name"修改为fastchat服务中的"api_base_url"
|
||
"api_key": "EMPTY"
|
||
}
|
||
}
|
||
|
||
VLLM_MODEL_DICT = {
|
||
'chatglm2-6b': "THUDM/chatglm-6b",
|
||
}
|
||
|
||
# or 若模型地址如下
|
||
model_dir: ~/codefuse-chatbot/llm_models/chatglm-6b
|
||
llm_model_dict = {
|
||
"chatglm-6b": {
|
||
"local_model_path": "chatglm-6b",
|
||
"api_base_url": "http://localhost:8888/v1", # "name"修改为fastchat服务中的"api_base_url"
|
||
"api_key": "EMPTY"
|
||
}
|
||
}
|
||
|
||
VLLM_MODEL_DICT = {
|
||
'chatglm2-6b': "chatglm-6b",
|
||
}
|
||
|
||
# 2、若不想移动相关模型到 ~/codefuse-chatbot/llm_models
|
||
# 同时删除 `模型路径重置` 以下的相关代码，具体见model_config.py
|
||
# 若模型地址如下
|
||
model_dir: ~/THUDM/chatglm-6b
|
||
# 参考配置如下
|
||
llm_model_dict = {
|
||
"chatglm-6b": {
|
||
"local_model_path": "~/THUDM/chatglm-6b",
|
||
"api_base_url": "http://localhost:8888/v1", # "name"修改为fastchat服务中的"api_base_url"
|
||
"api_key": "EMPTY"
|
||
}
|
||
}
|
||
|
||
VLLM_MODEL_DICT = {
|
||
'chatglm2-6b': "~/THUDM/chatglm-6b",
|
||
}
|
||
```
|
||
|
||
```bash
|
||
# 3、指定启动的模型服务，两者保持一致
|
||
LLM_MODEL = "gpt-3.5-turbo-16k"
|
||
LLM_MODELs = ["gpt-3.5-turbo-16k"]
|
||
```
|
||
|
||
```bash
|
||
# server_config.py配置修改， 若LLM_MODELS无多个模型配置不需要额外进行设置
|
||
# 修改server_config.py#FSCHAT_MODEL_WORKERS的配置
|
||
"model_name": {'host': DEFAULT_BIND_HOST, 'port': 20057}
|
||
```
|
||
|
||
|
||
|
||
<br>量化模型接入
|
||
|
||
```bash
|
||
# 若需要支撑codellama-34b-int4模型，需要给fastchat打一个补丁
|
||
cp examples/gptq.py ~/site-packages/fastchat/modules/gptq.py
|
||
|
||
# 若需要支撑qwen-72b-int4模型，需要给fastchat打一个补丁
|
||
cp examples/gptq.py ~/site-packages/fastchat/modules/gptq.py
|
||
# 量化需修改llm_api.py的配置
|
||
# dev_opsgpt/service/llm_api.py#559 取消注释 kwargs["gptq_wbits"] = 4
|
||
```
|
||
|
||
## 公开大模型接口接入
|
||
|
||
```bash
|
||
# model_config.py配置修改
|
||
# ONLINE_LLM_MODEL
|
||
# 其它接口开发来自于langchain-chatchat项目，缺少相关账号未经测试
|
||
|
||
# 指定启动的模型服务，两者保持一致
|
||
LLM_MODEL = "gpt-3.5-turbo-16k"
|
||
LLM_MODELs = ["gpt-3.5-turbo-16k"]
|
||
```
|
||
|
||
外部大模型接口接入示例
|
||
|
||
```bash
|
||
# 1、实现新的模型接入类
|
||
# 参考 ~/dev_opsgpt/service/model_workers/openai.py#ExampleWorker
|
||
# 实现do_chat函数即可使用LLM的能力
|
||
|
||
class XXWorker(ApiModelWorker):
|
||
def __init__(
|
||
self,
|
||
*,
|
||
controller_addr: str = None,
|
||
worker_addr: str = None,
|
||
model_names: List[str] = ["gpt-3.5-turbo"],
|
||
version: str = "gpt-3.5",
|
||
**kwargs,
|
||
):
|
||
kwargs.update(model_names=model_names, controller_addr=controller_addr, worker_addr=worker_addr)
|
||
kwargs.setdefault("context_len", 16384) #TODO 16K模型需要改成16384
|
||
super().__init__(**kwargs)
|
||
self.version = version
|
||
|
||
def do_chat(self, params: ApiChatParams) -> Dict:
|
||
'''
|
||
执行Chat的方法，默认使用模块里面的chat函数。
|
||
:params.messages : [
|
||
{"role": "user", "content": "hello"},
|
||
{"role": "assistant", "content": "hello"}
|
||
]
|
||
:params.xx: 详情见 ApiChatParams
|
||
要求返回形式：{"error_code": int, "text": str}
|
||
'''
|
||
return {"error_code": 500, "text": f"{self.model_names[0]}未实现chat功能"}
|
||
|
||
|
||
# 最后在 ~/dev_opsgpt/service/model_workers/__init__.py 中完成注册
|
||
# from .xx import XXWorker
|
||
|
||
# 2、通过已有模型接入类完成接入
|
||
# 或者直接使用已有的相关大模型类进行使用（缺少相关账号测试，欢迎大家测试后提PR）
|
||
```
|
||
|
||
|
||
```bash
|
||
# model_config.py#ONLINE_LLM_MODEL 配置修改
|
||
# 填写专属模型的 version、api_base_url、api_key、provider（与上述类名一致）
|
||
ONLINE_LLM_MODEL = {
|
||
# 线上模型。请在server_config中为每个在线API设置不同的端口
|
||
|
||
"openai-api": {
|
||
"model_name": "gpt-3.5-turbo",
|
||
"api_base_url": "https://api.openai.com/v1",
|
||
"api_key": "",
|
||
"openai_proxy": "",
|
||
},
|
||
"example": {
|
||
"version": "gpt-3.5", # 采用openai接口做示例
|
||
"api_base_url": "https://api.openai.com/v1",
|
||
"api_key": "",
|
||
"provider": "ExampleWorker",
|
||
},
|
||
}
|
||
```
|
||
|
||
## 启动大模型服务
|
||
```bash
|
||
# start llm-service（可选） 单独启动大模型服务
|
||
python dev_opsgpt/service/llm_api.py
|
||
```
|
||
|
||
```bash
|
||
# 启动测试
|
||
import openai
|
||
# openai.api_key = "EMPTY" # Not support yet
|
||
openai.api_base = "http://127.0.0.1:8888/v1"
|
||
|
||
# 选择你启动的模型
|
||
model = "example"
|
||
|
||
# create a chat completion
|
||
completion = openai.ChatCompletion.create(
|
||
model=model,
|
||
messages=[{"role": "user", "content": "Hello! What is your name? "}],
|
||
max_tokens=100,
|
||
)
|
||
# print the completion
|
||
print(completion.choices[0].message.content)
|
||
|
||
# 正确输出后则确认LLM可正常接入
|
||
```
|
||
|
||
|
||
|
||
or
|
||
|
||
```bash
|
||
# model_config.py#USE_FASTCHAT 判断是否进行fastchat接入本地模型
|
||
USE_FASTCHAT = "gpt" not in LLM_MODEL
|
||
python start.py #224 自动执行 python service/llm_api.py
|
||
``` |