当前位置:   article > 正文

本地部署 闻达:一个LLM调用平台 VIP版_闻达 github

闻达 github

闻达:一个LLM调用平台。目标为针对特定环境的高效内容生成,同时考虑个人和中小企业的计算资源局限性,以及知识安全和私密性问题

1. 什么是 闻达

本项目设计目标为实现针对特定环境的高效内容生成,同时考虑个人和中小企业的计算资源局限性,以及知识安全和私密性问题。为达目标,平台化集成了以下能力:

  1. 知识库:支持对接本地离线向量库本地搜索引擎、在线搜索引擎等。
  2. 多种大语言模型:目前支持离线部署模型有chatGLM-6BchatRWKVllama系列(不推荐中文用户)moss(不推荐)baichuan(不推荐)Aquila-7B,在线API访问openai apichatGLM-130b api
  3. Auto脚本:通过开发插件形式的JavaScript脚本,为平台附件功能,实现包括但不限于自定义对话流程、访问外部API、在线切换LoRA模型。
  4. 其他实用化所需能力:对话历史管理、内网部署、多用户同时使用等。

2. 闻达 Github 地址

https://github.com/wenda-LLM/wenda.git

3. 安装 Miniconda3

下载 Conda 安装脚本,

wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
  • 1

运行安装脚本,

bash Miniconda3-latest-Linux-x86_64.sh
  • 1

按提示操作。当提示是否初始化 Conda 时,输入 “yes”,

在这里插入图片描述

安装完成后,关闭当前终端并打开新终端,这将激活 Conda,

sudo su - root
  • 1

更新 Conda 至最新版本,

conda update conda
  • 1

在这里插入图片描述
在这里插入图片描述

4. 创建虚拟环境

conda create -n wenda python==3.10.6
conda activate wenda
  • 1
  • 2

5. 部署 闻达

克隆代码,

git clone https://github.com/wenda-LLM/wenda.git; cd wenda
  • 1

创建存放模型的目录,

mkdir model
  • 1

创建 txt 的目录,

mkdir txt
  • 1

创建配置文件,

cp example.config.yml config.yml
  • 1

安装库,

pip install -r requirements/requirements.txt
  • 1

验证 cuda 是否可用,

python -c "import torch; print(torch.cuda.is_available());"

--- 输出 True 表示 cuda 可用
True
---
  • 1
  • 2
  • 3
  • 4
  • 5

下载 m3e-base 模型,

git lfs install
cd model/
git clone https://huggingface.co/moka-ai/m3e-base
cd ..
  • 1
  • 2
  • 3
  • 4

6. 部署 RWKV-4-Raven-14B 模型

访问 https://huggingface.co/BlinkDL/rwkv-4-raven/tree/main,下载最新版 RWKV-4-Raven-14B 模型,

在这里插入图片描述
将下载后的模型拷贝/移动到 model 目录,

ls model

--- 输出中有 RWKV-4-Raven-14B 模型文件
RWKV-4-Raven-14B-v12-Eng98%-Other2%-20230523-ctx8192.pth
---
  • 1
  • 2
  • 3
  • 4
  • 5

修改配置文件中 RWKV-4-Raven-14B 模型的 path,

vi config.yml
  • 1
--- 将各个 model 的 path 修改正确
llm_models:
  rwkv:
     path: "model/RWKV-4-Raven-14B-v12-Eng98%-Other2%-20230523-ctx8192.pth"      #rwkv模型位置"
---
  • 1
  • 2
  • 3
  • 4
  • 5

启动 RWKV,

./run_rwkv.sh
  • 1

使用浏览器打开 http://127.0.0.1:17860/ 进行访问,问它几个问题,大家可以通过截图体验一下它的回答水平怎么样。

在这里插入图片描述

7. 部署 AquilaChat-7B 模型

创建 run_aquila.sh 启动脚本文件,

cat << "EOF" > run_aquila.sh
#!/bin/bash
PYTHON=""
# python程序位置,可搭配一键包或是省去每次切换环境

while true
do
    if [ -z "$PYTHON" ]; then
        python wenda.py -t aquila
    else
        $PYTHON wenda.py -t aquila
    fi
sleep 1
done
EOF
  • 1
  • 2
  • 3
  • 4
  • 5
  • 6
  • 7
  • 8
  • 9
  • 10
  • 11
  • 12
  • 13
  • 14
  • 15
chmod +x run_aquila.sh
  • 1

启动 AquilaChat,

./run_aquila.sh
  • 1

使用浏览器打开 http://127.0.0.1:17860/ 进行访问,问它几个问题,大家可以通过截图体验一下它的回答水平怎么样。

在这里插入图片描述

8. 部署 baichuan-7B 模型

下载 baichuan-7B 模型,

git lfs install
cd model/
git clone https://huggingface.co/baichuan-inc/baichuan-7B
cd ..
  • 1
  • 2
  • 3
  • 4

创建 run_baichuan.sh 启动脚本文件,

cat << "EOF" > run_baichuan.sh
#!/bin/bash
PYTHON=""
# python程序位置,可搭配一键包或是省去每次切换环境

while true
do
    if [ -z "$PYTHON" ]; then
        python wenda.py -t baichuan
    else
        $PYTHON wenda.py -t baichuan
    fi
sleep 1
done
EOF
  • 1
  • 2
  • 3
  • 4
  • 5
  • 6
  • 7
  • 8
  • 9
  • 10
  • 11
  • 12
  • 13
  • 14
  • 15
chmod +x run_baichuan.sh
  • 1

启动 baichuan,

./run_baichuan.sh
  • 1

使用浏览器打开 http://127.0.0.1:17860/ 进行访问,问它几个问题,大家可以通过截图体验一下它的回答水平怎么样。

在这里插入图片描述

9. 部署 baichuan-7B-chat 模型

下载 baichuan-7B-chat 模型,

git lfs install
cd model/
git clone https://huggingface.co/csdc-atl/baichuan-7B-cha
  • 1
  • 2
声明:本文内容由网友自发贡献,不代表【wpsshop博客】立场,版权归原作者所有,本站不承担相应法律责任。如您发现有侵权的内容,请联系我们。转载请注明出处:https://www.wpsshop.cn/w/Cpp五条/article/detail/72730
推荐阅读
相关标签
  

闽ICP备14008679号