欢迎关注我的CSDN：https://spike.blog.csdn.net/
本文地址：https://spike.blog.csdn.net/article/details/148421901

免责声明：本文来源于个人知识与公开资料，仅用于学术交流，欢迎讨论，不支持转载。

---

建议使用 Docker 配置 PyTorch 研发环境，原因是部分机器配置差异较大，而且环境各不相同，导致安装到最后仍然无法启动训练任务，浪费大量时间。建议直接使用 Docker + Conda(Mamba) 环境构建虚拟环境，即可支持多数任务。

1. 网络代理

常用的 GitHub 下载较慢，建议使用代理，提速。

使用特定的网络代理，即：

export https_proxy=http://xxx:80
export http_proxy=http://xxx:80

unset https_proxy http_proxy

xxx 是 IP 地址。

或 使用在线的免费代理，即：https://ghproxy.link/

# https://ghfast.top
git clone https://ghfast.top/https://github.com/hiyouga/LLaMA-Factory.git   # 示例

注意：免费代理可能失效，需要实时查看。

Huggingface 环境，参考：https://hf-mirror.com/：

export HF_ENDPOINT=https://hf-mirror.com

2. 环境变量

打印系统环境变量：

printenv

配置大模型相关的环境变量，写入 ~/.bashrc 如下：

export WORK_DIR="xxx"
export TORCH_HOME="$WORK_DIR/torch_home/"
export HF_HOME="$WORK_DIR/huggingface/"
export HUGGINGFACE_TOKEN="xxx"
export MODELSCOPE_CACHE="$WORK_DIR/modelscope_models/"
export MODELSCOPE_API_TOKEN="xxx"
export CUDA_HOME="/usr/local/cuda"
export OMP_NUM_THREADS=64

3. Docker

建议使用 Nvidia 的镜像，其中包含默认的配置与环境：https://docker.aityp.com/r/docker.io/nvcr.io/nvidia/pytorch

拉取 Docker 镜像(国内代理)，建议使用 24.12-py3 版本，不要使用最新版本，兼容异常：

docker pull swr.cn-north-4.myhuaweicloud.com/ddn-k8s/docker.io/nvcr.io/nvidia/pytorch:24.12-py3

启动 Docker 的标准模版，即：

docker run -itd \
--name [your name] \
--gpus all \
--shm-size=128g \
--memory=256g \
--cpus=64 \
--restart=unless-stopped \
--ipc=host \
--ulimit memlock=-1 \
--ulimit stack=67108864 \
-v [your path]:[your path] \
-e NVIDIA_DRIVER_CAPABILITIES=compute,utility \
--privileged \
--network host \
swr.cn-north-4.myhuaweicloud.com/ddn-k8s/docker.io/nvcr.io/nvidia/pytorch:24.12-py3 \
/bin/bash

4. 虚拟环境

建议，使用 Conda 或 Mamba，以 Mamba 为例：

"${SHELL}" <(curl -L micro.mamba.pm/install.sh)

手动下载模式，直接下载 micro.mamba.pm/install.sh，即 GitHub 路径，同时 替换代理 https://ghfast.top/。

配置 pip 源：

# docker 优先级
rm -rf /usr/pip.conf
rm -rf /root/.config/pip/pip.conf
rm -rf /etc/pip.conf
rm -rf /etc/xdg/pip/pip.conf

# 配置其他源
mkdir ~/.pip
vim ~/.pip/pip.conf

[global]
no-cache-dir = true
index-url = http://mirrors.aliyun.com/pypi/simple/
trusted-host = mirrors.aliyun.com

安装 torch_def 环境：

micromamba create -n torch_def python=3.11
pip3 install torch torchvision torchaudio --timeout=100

下载速度较慢，避免超时 --timeout=100

验证 PyTorch 环境：

import torch
print(torch.__version__)  			# 2.7.0+cu126
print(torch.cuda.is_available())  	# True

下载相关 Python 包：

pip install datasets accelerate bitsandbytes peft swanlab sentencepiece trl deepspeed modelscope
pip install -U "huggingface_hub[cli]"

配置下载 Huggingface 模型与数据集，参考：

huggingface-cli download Qwen/Qwen3-8B --local-dir Qwen/Qwen3-8B
huggingface-cli download --repo-type dataset FreedomIntelligence/medical-o1-reasoning-SFT --local-dir FreedomIntelligence/medical-o1-reasoning-SFT

参考数据集：FreedomIntelligence/medical-o1-reasoning-SFT

其他

已优化的 Mamba 安装文件 mamba_install.sh，如下：

#!/bin/sh

set -eu

# Detect the shell from which the script was called
parent=$(ps -o comm $PPID |tail -1)
parent=${parent#-}  # remove the leading dash that login shells have
case "$parent" in
  # shells supported by `micromamba shell init`
  bash|fish|xonsh|zsh)
    shell=$parent
    ;;
  *)
    # use the login shell (basename of $SHELL) as a fallback
    shell=${SHELL##*/}
    ;;
esac

# Parsing arguments
if [ -t 0 ] ; then
  printf "Micromamba binary folder? [~/.local/bin] "
  read BIN_FOLDER
  printf "Init shell ($shell)? [Y/n] "
  read INIT_YES
  printf "Configure conda-forge? [Y/n] "
  read CONDA_FORGE_YES
fi

# Fallbacks
BIN_FOLDER="${BIN_FOLDER:-${HOME}/.local/bin}"
INIT_YES="${INIT_YES:-yes}"
CONDA_FORGE_YES="${CONDA_FORGE_YES:-yes}"

# Prefix location is relevant only if we want to call `micromamba shell init`
case "$INIT_YES" in
  y|Y|yes)
    if [ -t 0 ]; then
      printf "Prefix location? [~/micromamba] "
      read PREFIX_LOCATION
    fi
    ;;
esac
PREFIX_LOCATION="${PREFIX_LOCATION:-${HOME}/micromamba}"

# Computing artifact location
case "$(uname)" in
  Linux)
    PLATFORM="linux" ;;
  Darwin)
    PLATFORM="osx" ;;
  *NT*)
    PLATFORM="win" ;;
esac

ARCH="$(uname -m)"
case "$ARCH" in
  aarch64|ppc64le|arm64)
      ;;  # pass
  *)
    ARCH="64" ;;
esac

case "$PLATFORM-$ARCH" in
  linux-aarch64|linux-ppc64le|linux-64|osx-arm64|osx-64|win-64)
      ;;  # pass
  *)
    echo "Failed to detect your OS" >&2
    exit 1
    ;;
esac

if [ "${VERSION:-}" = "" ]; then
  RELEASE_URL="https://ghfast.top/https://github.com/mamba-org/micromamba-releases/releases/latest/download/micromamba-${PLATFORM}-${ARCH}"
else
  RELEASE_URL="https://ghfast.top/https://github.com/mamba-org/micromamba-releases/releases/download/${VERSION}/micromamba-${PLATFORM}-${ARCH}"
fi


# Downloading artifact
mkdir -p "${BIN_FOLDER}"
if hash curl >/dev/null 2>&1; then
  curl "${RELEASE_URL}" -o "${BIN_FOLDER}/micromamba" -fsSL --compressed ${CURL_OPTS:-}
elif hash wget >/dev/null 2>&1; then
  wget ${WGET_OPTS:-} -qO "${BIN_FOLDER}/micromamba" "${RELEASE_URL}"
else
  echo "Neither curl nor wget was found" >&2
  exit 1
fi
chmod +x "${BIN_FOLDER}/micromamba"


# Initializing shell
case "$INIT_YES" in
  y|Y|yes)
    case $("${BIN_FOLDER}/micromamba" --version) in
      1.*|0.*)
        shell_arg=-s
        prefix_arg=-p
        ;;
      *)
        shell_arg=--shell
        prefix_arg=--root-prefix
        ;;
    esac
    "${BIN_FOLDER}/micromamba" shell init $shell_arg "$shell" $prefix_arg "$PREFIX_LOCATION"

    echo "Please restart your shell to activate micromamba or run the following:\n"
    echo "  source ~/.bashrc (or ~/.zshrc, ~/.xonshrc, ~/.config/fish/config.fish, ...)"
    ;;
  *)
    echo "You can initialize your shell later by running:"
    echo "  micromamba shell init"
    ;;
esac


# Initializing conda-forge
case "$CONDA_FORGE_YES" in
  y|Y|yes)
    "${BIN_FOLDER}/micromamba" config append channels conda-forge
    "${BIN_FOLDER}/micromamba" config append channels nodefaults
    "${BIN_FOLDER}/micromamba" config set channel_priority strict
    ;;
esac