4 llama3仿openai api实验 本地服务器 #7

New Issue

Open

opened 2024-10-20 21:41:30 +08:00 by 12390900721cs · 0 comments

12390900721cs commented 2024-10-20 21:41:30 +08:00

Copy Link

1. 部署Chinese-LLaMA-Alpaca-3项目；
2. 封装一个私有的兼容openai api的大模型接口；
3. 使用ChatGPTNextWeb开源工具调用此接口。

本文默认您在操作系统的用户名为：ganjialing，如果您使用的是其他用户名，请更改涉及到用户名的地址。

环境准备

源码下载

cd ~/ && wget https://file.huishiwei.top/Chinese-LLaMA-Alpaca-3-3.0.tar.gz
tar -xvf Chinese-LLaMA-Alpaca-3-3.0.tar.gz

安装miniconda

cd ~/ && wget https://repo.anaconda.com/miniconda/Miniconda3-py38_23.5.2-0-Linux-x86_64.sh
bash Miniconda3-py38_23.5.2-0-Linux-x86_64.sh

创建并激活虚拟环境

conda create -n chinese_llama_alpaca_3 python=3.8.17 pip -y
conda activate chinese_llama_alpaca_3

下载模型

pip install modelscope -i https://mirrors.aliyun.com/pypi/simple
modelscope download --model ChineseAlpacaGroup/llama-3-chinese-8b-instruct-v3

查看模型存储位置

ls -alhrt ~/.cache/modelscope/hub/ChineseAlpacaGroup/llama-3-chinese-8b-instruct-v3

开源版openai接口启动

中文llama3的开源版本实现在以下目录：/home/ganjialing/Chinese-LLaMA-Alpaca-3-3.0/scripts/oai_api_demo，下面分别是GPU和CPU版本的启动流程。

脚本bug修复

打开/home/llm_course/Chinese-LLaMA-Alpaca-3-3.0/scripts/oai_api_demo/openai_api_server.py文件，找到如下内容：

def stream_predict(
    input,
    max_new_tokens=1024,
    top_p=0.9,
    temperature=0.2,
    top_k=40,
    num_beams=4,
    repetition_penalty=1.1,
    do_sample=True,
    model_id="llama-3-chinese",
    **kwargs,
):
    choice_data = ChatCompletionResponseStreamChoice(
        index=0, delta=DeltaMessage(role="assistant"), finish_reason=None
    )
    chunk = ChatCompletionResponse(
        model=model_id,
        choices=[choice_data],
        object="chat.completion.chunk",
    )
    yield "{}".format(chunk.json(exclude_unset=True, ensure_ascii=False))

    if isinstance(input, str):
        prompt = generate_completion_prompt(input)
    else:
        prompt = generate_chat_prompt(input)
    inputs = tokenizer(prompt, return_tensors="pt")
    input_ids = inputs["input_ids"].to(device)
    generation_config = GenerationConfig(
        temperature=temperature,
        top_p=top_p,
        top_k=top_k,
        num_beams=num_beams,
        do_sample=do_sample,
        **kwargs,
    )

    streamer = TextIteratorStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
    generation_kwargs = dict(
        streamer=streamer,
        input_ids=input_ids,
        generation_config=generation_config,
        return_dict_in_generate=True,
        output_scores=False,
        max_new_tokens=max_new_tokens,
        repetition_penalty=float(repetition_penalty)
      )

将generation_kwargs改为如下值：

generation_kwargs = dict(
    streamer=streamer,
    input_ids=input_ids,
    generation_config=generation_config,
    return_dict_in_generate=True,
    output_scores=False,
    max_new_tokens=max_new_tokens,
    repetition_penalty=float(repetition_penalty),
    pad_token_id=tokenizer.eos_token_id, # 新添加的参数
    eos_token_id=[tokenizer.eos_token_id, tokenizer.convert_tokens_to_ids("<|eot_id|>")] # 新添加的参数
)

加入上面的参数主要是为了兼容llama3特有的停止token，不然流式接口返回的内容会不断的自动重复，不停止。

GPU版本接口启动

备份脚本

使用如下命令备份/home/ganjialing/Chinese-LLaMA-Alpaca-3-3.0/requirements.txt文件：

mv /home/ganjialing/Chinese-LLaMA-Alpaca-3-3.0/requirements.txt /home/ganjialing/Chinese-LLaMA-Alpaca-3-3.0/requirements.bk.tx

安装依赖

通过如下命令创建新的requirements.txt：

cat <<EOF > /home/ganjialing/Chinese-LLaMA-Alpaca-3-3.0/requirements.txt
accelerate==0.30.0
aiohttp==3.9.5
aiosignal==1.3.1
annotated-types==0.6.0
anyio==4.3.0
async-timeout==4.0.3
attrs==23.2.0
bitsandbytes==0.43.1
certifi==2024.2.2
charset-normalizer==3.3.2
click==8.1.7
datasets==2.20.0
deepspeed==0.13.1
dill==0.3.7
dnspython==2.6.1
einops==0.8.0
email_validator==2.1.1
exceptiongroup==1.2.1
fastapi==0.109.2
fastapi-cli==0.0.3
filelock==3.14.0
frozenlist==1.4.1
fsspec==2023.10.0
h11==0.14.0
hjson==3.1.0
httpcore==1.0.5
httptools==0.6.1
httpx==0.27.0
huggingface-hub==0.23.3
idna==3.7
Jinja2==3.1.4
joblib==1.4.2
markdown-it-py==3.0.0
MarkupSafe==2.1.5
mdurl==0.1.2
modelscope==1.17.1
mpmath==1.3.0
multidict==6.0.5
multiprocess==0.70.15
networkx==3.1
ninja==1.11.1.1
numpy==1.24.4
nvidia-cublas-cu12==12.1.3.1
nvidia-cuda-cupti-cu12==12.1.105
nvidia-cuda-nvrtc-cu12==12.1.105
nvidia-cuda-runtime-cu12==12.1.105
nvidia-cudnn-cu12==8.9.2.26
nvidia-cufft-cu12==11.0.2.54
nvidia-curand-cu12==10.3.2.106
nvidia-cusolver-cu12==11.4.5.107
nvidia-cusparse-cu12==12.1.0.106
nvidia-nccl-cu12==2.18.1
nvidia-nvjitlink-cu12==12.4.127
nvidia-nvtx-cu12==12.1.105
orjson==3.10.3
packaging==24.0
pandas==2.0.3
peft==0.7.1
psutil==5.9.8
py-cpuinfo==9.0.0
pyarrow==16.0.0
pyarrow-hotfix==0.6
pydantic==1.10.11
pydantic_core==2.18.2
Pygments==2.18.0
pynvml==11.5.0
python-dateutil==2.9.0.post0
python-decouple==3.8
python-dotenv==1.0.1
python-multipart==0.0.9
pytz==2024.1
PyYAML==6.0.1
regex==2024.4.28
requests==2.32.3
rich==13.7.1
safetensors==0.4.3
scikit-learn==1.3.2
scipy==1.10.1
shellingham==1.5.4
shortuuid==1.0.13
six==1.16.0
sniffio==1.3.1
sse-starlette==2.1.0
starlette==0.36.3
sympy==1.12
threadpoolctl==3.5.0
tokenizers==0.19.1
torch==2.1.2
tqdm==4.66.4
transformers==4.41.2
triton==2.1.0
typer==0.12.3
typing_extensions==4.11.0
tzdata==2024.1
ujson==5.9.0
urllib3==2.2.1
uvicorn==0.29.0
uvloop==0.19.0
watchfiles==0.21.0
websockets==12.0
xxhash==3.4.1
yarl==1.9.4
EOF

安装依赖：

pip install -r requirements.txt -i https://mirrors.aliyun.com/pypi/simple

启动大模型模型

python /home/ganjialing/Chinese-LLaMA-Alpaca-3-3.0/scripts/oai_api_demo/openai_api_server.py --gpus 0 --base_model /home/ganjialing/.cache/modelscope/hub/ChineseAlpacaGroup/llama-3-chinese-8b-instruct-v3

测试效果

安装ChatNextWeb

我们使用ChatGPTNextWeb工具测试我们的接口，下载：

Windows：https://github.com/ChatGPTNextWeb/ChatGPT-Next-Web/releases/download/v2.14.2/NextChat_2.14.2_x64-setup.exe

接入大模型

设置

设置接口地址，若为本地则为localhost+端口号，服务器则需要替换；

API_KEY随便设置就好。

这里把ChatGPT提示信息去掉；

设置完后点这里关闭；

点这里选择模型；

选择llama-chinese；

测试可通。

CPU版本接口启动

创建CPU版本专用的虚拟环境

创造并启动conda环境：

conda create -n chinese_llama_alpaca_3_cpu python=3.8.17 pip -y
conda activate chinese_llama_alpaca_3_cpu

安装依赖：

pip3 install torch==2.3.0 --index-url https://download.pytorch.org/whl/cpu
pip3 install fastapi==0.111.0 peft==0.7.1 pydantic==1.10.11 pydantic_core==2.18.2 shortuuid==1.0.13 sse-starlette==2.1.0 starlette==0.37.2 transformers==4.41.2 -i https://mirrors.aliyun.com/pypi/simple

利用CPU启动大模型服务

python openai_api_server.py --only_cpu --base_model /home/ganjialing/.cache/modelscope/hub/ChineseAlpacaGroup/llama-3-chinese-8b-instruct-v3

测试效果

与CPU版本相同。

cpu还是慢一些的。

1. 部署<font style="color:black;">Chinese-LLaMA-Alpaca-3项目；</font> 2. <font style="color:black;">封装一个私有的兼容openai api的大模型接口；</font> 3. <font style="color:black;">使用ChatGPTNextWeb开源工具调用此接口。</font> <font style="color:black;"></font> <font style="color:black;">本文默认您在操作系统的用户名为：ganjialing，如果您使用的是其他用户名，请更改涉及到用户名的地址。</font> <font style="color:black;"></font> ## 环境准备 ### 源码下载 ```plain cd ~/ && wget https://file.huishiwei.top/Chinese-LLaMA-Alpaca-3-3.0.tar.gz tar -xvf Chinese-LLaMA-Alpaca-3-3.0.tar.gz ```  ### 安装miniconda ```plain cd ~/ && wget https://repo.anaconda.com/miniconda/Miniconda3-py38_23.5.2-0-Linux-x86_64.sh bash Miniconda3-py38_23.5.2-0-Linux-x86_64.sh ```   ### 创建并激活虚拟环境 ```plain conda create -n chinese_llama_alpaca_3 python=3.8.17 pip -y conda activate chinese_llama_alpaca_3 ```   ### 下载模型 ```plain pip install modelscope -i https://mirrors.aliyun.com/pypi/simple modelscope download --model ChineseAlpacaGroup/llama-3-chinese-8b-instruct-v3 ```   ### 查看模型存储位置 ```plain ls -alhrt ~/.cache/modelscope/hub/ChineseAlpacaGroup/llama-3-chinese-8b-instruct-v3 ```  ## 开源版openai接口启动 <font style="color:black;">中文llama3的开源版本实现在以下目录：/home/ganjialing/Chinese-LLaMA-Alpaca-3-3.0/scripts/oai_api_demo，下面分别是GPU和CPU版本的启动流程。</font> <font style="color:black;"></font> ### <font style="color:black;">脚本bug修复</font> 打<font style="color:black;">开/home/llm_course/Chinese-LLaMA-Alpaca-3-3.0/scripts/oai_api_demo/openai_api_server.py文件，找到如下内容：</font> ```plain def stream_predict( input, max_new_tokens=1024, top_p=0.9, temperature=0.2, top_k=40, num_beams=4, repetition_penalty=1.1, do_sample=True, model_id="llama-3-chinese", **kwargs, ): choice_data = ChatCompletionResponseStreamChoice( index=0, delta=DeltaMessage(role="assistant"), finish_reason=None ) chunk = ChatCompletionResponse( model=model_id, choices=[choice_data], object="chat.completion.chunk", ) yield "{}".format(chunk.json(exclude_unset=True, ensure_ascii=False)) if isinstance(input, str): prompt = generate_completion_prompt(input) else: prompt = generate_chat_prompt(input) inputs = tokenizer(prompt, return_tensors="pt") input_ids = inputs["input_ids"].to(device) generation_config = GenerationConfig( temperature=temperature, top_p=top_p, top_k=top_k, num_beams=num_beams, do_sample=do_sample, **kwargs, ) streamer = TextIteratorStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True) generation_kwargs = dict( streamer=streamer, input_ids=input_ids, generation_config=generation_config, return_dict_in_generate=True, output_scores=False, max_new_tokens=max_new_tokens, repetition_penalty=float(repetition_penalty) ) ``` <font style="color:black;"></font> <font style="color:black;">将generation_kwargs改为如下值：</font> ```plain generation_kwargs = dict( streamer=streamer, input_ids=input_ids, generation_config=generation_config, return_dict_in_generate=True, output_scores=False, max_new_tokens=max_new_tokens, repetition_penalty=float(repetition_penalty), pad_token_id=tokenizer.eos_token_id, # 新添加的参数 eos_token_id=[tokenizer.eos_token_id, tokenizer.convert_tokens_to_ids("<|eot_id|>")] # 新添加的参数 ) ``` <font style="color:black;">加入上面的参数主要是为了兼容llama3特有的停止token，不然流式接口返回的内容会不断的自动重复，不停止。</font>  ### GPU版本接口启动 #### 备份脚本 <font style="color:black;">使用如下命令备份/home/ganjialing/Chinese-LLaMA-Alpaca-3-3.0/requirements.txt文件：</font> ```plain mv /home/ganjialing/Chinese-LLaMA-Alpaca-3-3.0/requirements.txt /home/ganjialing/Chinese-LLaMA-Alpaca-3-3.0/requirements.bk.tx ```  #### 安装依赖 <font style="color:black;">通过如下命令创建新的requirements.txt：</font> ```plain cat <<EOF > /home/ganjialing/Chinese-LLaMA-Alpaca-3-3.0/requirements.txt accelerate==0.30.0 aiohttp==3.9.5 aiosignal==1.3.1 annotated-types==0.6.0 anyio==4.3.0 async-timeout==4.0.3 attrs==23.2.0 bitsandbytes==0.43.1 certifi==2024.2.2 charset-normalizer==3.3.2 click==8.1.7 datasets==2.20.0 deepspeed==0.13.1 dill==0.3.7 dnspython==2.6.1 einops==0.8.0 email_validator==2.1.1 exceptiongroup==1.2.1 fastapi==0.109.2 fastapi-cli==0.0.3 filelock==3.14.0 frozenlist==1.4.1 fsspec==2023.10.0 h11==0.14.0 hjson==3.1.0 httpcore==1.0.5 httptools==0.6.1 httpx==0.27.0 huggingface-hub==0.23.3 idna==3.7 Jinja2==3.1.4 joblib==1.4.2 markdown-it-py==3.0.0 MarkupSafe==2.1.5 mdurl==0.1.2 modelscope==1.17.1 mpmath==1.3.0 multidict==6.0.5 multiprocess==0.70.15 networkx==3.1 ninja==1.11.1.1 numpy==1.24.4 nvidia-cublas-cu12==12.1.3.1 nvidia-cuda-cupti-cu12==12.1.105 nvidia-cuda-nvrtc-cu12==12.1.105 nvidia-cuda-runtime-cu12==12.1.105 nvidia-cudnn-cu12==8.9.2.26 nvidia-cufft-cu12==11.0.2.54 nvidia-curand-cu12==10.3.2.106 nvidia-cusolver-cu12==11.4.5.107 nvidia-cusparse-cu12==12.1.0.106 nvidia-nccl-cu12==2.18.1 nvidia-nvjitlink-cu12==12.4.127 nvidia-nvtx-cu12==12.1.105 orjson==3.10.3 packaging==24.0 pandas==2.0.3 peft==0.7.1 psutil==5.9.8 py-cpuinfo==9.0.0 pyarrow==16.0.0 pyarrow-hotfix==0.6 pydantic==1.10.11 pydantic_core==2.18.2 Pygments==2.18.0 pynvml==11.5.0 python-dateutil==2.9.0.post0 python-decouple==3.8 python-dotenv==1.0.1 python-multipart==0.0.9 pytz==2024.1 PyYAML==6.0.1 regex==2024.4.28 requests==2.32.3 rich==13.7.1 safetensors==0.4.3 scikit-learn==1.3.2 scipy==1.10.1 shellingham==1.5.4 shortuuid==1.0.13 six==1.16.0 sniffio==1.3.1 sse-starlette==2.1.0 starlette==0.36.3 sympy==1.12 threadpoolctl==3.5.0 tokenizers==0.19.1 torch==2.1.2 tqdm==4.66.4 transformers==4.41.2 triton==2.1.0 typer==0.12.3 typing_extensions==4.11.0 tzdata==2024.1 ujson==5.9.0 urllib3==2.2.1 uvicorn==0.29.0 uvloop==0.19.0 watchfiles==0.21.0 websockets==12.0 xxhash==3.4.1 yarl==1.9.4 EOF ```  安装依赖： ```plain pip install -r requirements.txt -i https://mirrors.aliyun.com/pypi/simple ```  #### 启动大模型模型 ```plain python /home/ganjialing/Chinese-LLaMA-Alpaca-3-3.0/scripts/oai_api_demo/openai_api_server.py --gpus 0 --base_model /home/ganjialing/.cache/modelscope/hub/ChineseAlpacaGroup/llama-3-chinese-8b-instruct-v3 ```  #### 测试效果 ##### <font style="color:black;">安装ChatNextWeb</font> <font style="color:black;">我们使用ChatGPTNextWeb工具测试我们的接口，下载：</font> <font style="color:black;">Windows：</font>[https://github.com/ChatGPTNextWeb/ChatGPT-Next-Web/releases/download/v2.14.2/NextChat_2.14.2_x64-setup.exe](https://github.com/ChatGPTNextWeb/ChatGPT-Next-Web/releases/download/v2.14.2/NextChat_2.14.2_x64-setup.exe) ##### 接入大模型 设置  设置接口地址，若为本地则为localhost+端口号，服务器则需要替换； API_KEY随便设置就好。  这里把ChatGPT提示信息去掉；  设置完后点这里关闭；  点这里选择模型；  选择llama-chinese；  测试可通。  ### CPU版本接口启动 #### 创建CPU版本专用的虚拟环境 创造并启动conda环境： ```plain conda create -n chinese_llama_alpaca_3_cpu python=3.8.17 pip -y conda activate chinese_llama_alpaca_3_cpu ```   安装依赖： ```plain pip3 install torch==2.3.0 --index-url https://download.pytorch.org/whl/cpu pip3 install fastapi==0.111.0 peft==0.7.1 pydantic==1.10.11 pydantic_core==2.18.2 shortuuid==1.0.13 sse-starlette==2.1.0 starlette==0.37.2 transformers==4.41.2 -i https://mirrors.aliyun.com/pypi/simple ```   #### 利用CPU启动大模型服务 ```plain python openai_api_server.py --only_cpu --base_model /home/ganjialing/.cache/modelscope/hub/ChineseAlpacaGroup/llama-3-chinese-8b-instruct-v3 ```  #### 测试效果 与CPU版本相同。  cpu还是慢一些的。

Sign in to join this conversation.

No Branch/Tag Specified

Branches Tags

main

No results found.

Labels

Clear labels

No items

No Label

Milestone

Clear milestone

No items

No Milestone

Projects

Clear projects

No project

Assignees

Clear assignees

No Assignees

1 Participants

Notifications

Subscribe

Due Date

The due date is invalid or out of range. Please use the format 'yyyy-mm-dd'.

No due date set.

Dependencies

No dependencies set.

Reference: HswOAuth/llm_share#7

No description provided.