单击链接下载代码仓压缩包并上传至服务器任意目录并解压，例如“/home”。

进入代码目录修改代码文件。

修改"FastChat/pyproject.toml"文件。

cd FastChat-76f0424d1add61aadc8e5bdeed5ebe540f266ba3
vi pyproject.toml

修改前：

[project]
name = "fschat"
version = "0.2.2"
description = "An open platform for training,serving,and evaluating large language model based chatbots."
readme = "README.md"
requires-python = ">=3.8"             
classifiers = [
    "Programming Language :: Python :: 3",
    "License :: OSI Approved :: Apache Software License",
]
dependencies = [
    "accelerate", "fastAPI", "gradio==3.23", "markdown2[all]", "numpy",
    "prompt_toolkit>=3.0.0", "requests", "rich>=10.0.0", "sentencepiece",
    "shortuuid", "transformers>=4.28.0", "tokenizers>=0.12.1", "torch",
    "uvicorn", "wandb", "httpx", "shortuuid", "pydantic",
]

修改后（python 一般使用3.7较多，并固定transformers版本）：

[project]
name = "fschat"
version = "0.2.2"
description = "An open platform for training, serving, and evaluating large language model based chatbots."
readme = "README.md"
requires-python = ">=3.7"          # 当前 NPU python3.7使用较多
classifiers = [
    "Programming Language :: Python :: 3",
    "License :: OSI Approved :: Apache Software License",
]
dependencies = [
    "accelerate", "fastAPI", "gradio==3.23", "markdown2[all]", "numpy",
    "prompt_toolkit>=3.0.0", "requests", "rich>=10.0.0", "sentencepiece",
    "shortuuid", "transformers==4.28.1", "tokenizers>=0.12.1",    # 锁定修改版本
    "uvicorn", "httpx", "shortuuid", "pydantic",
]

执行以下命令安装依赖与模型套件。

pip3 install einops sympy regex decorator scipy setuptools_scm prompt_toolkit
pip3 install -e . -i https://pypi.tuna.tsinghua.edu.cn/simple
cd ..

安装deepspeed模型套件。

pip3 install deepspeed==0.9.2 
git clone https://gitee.com/ascend/DeepSpeed.git
cd DeepSpeed
pip3 install .  # 安装deepspeed_npu
cd ..

迁移前准备