Intel Extension for PyTorch兼容Anaconda吗？如何解决安装失败问题？

问题 #

我有一块不兼容CUDA的Intel GPU，只能在CPU上训练小型网络，耗时很长。尝试在Anaconda环境中安装PyTorch的intel_extension_for_pytorch包时，出现以下错误：

Collecting package metadata (current_repodata.json): ...working... done
Solving environment: ...working... failed with initial frozen solve. Retrying with flexible solve.
Collecting package metadata (repodata.json): ...working... done
Solving environment: ...working... failed with initial frozen solve. Retrying with flexible solve.

Note: you may need to restart the kernel to use updated packages.

PackagesNotFoundError: The following packages are not available from current channels:

  - intel_extension_for_pytorch

Current channels:

  - https://repo.anaconda.com/pkgs/main/win-64
  - https://repo.anaconda.com/pkgs/main/noarch
  - https://repo.anaconda.com/pkgs/r/win-64
  - https://repo.anaconda.com/pkgs/r/noarch
  - https://repo.anaconda.com/pkgs/msys2/win-64
  - https://repo.anaconda.com/pkgs/msys2/noarch

To search for alternate channels that may provide the conda package you're
looking for, navigate to

    https://anaconda.org

查阅该包的ReadMe和安装指南后，发现没有conda安装说明，仅提及pip安装方式，请问如何在Anaconda环境中成功安装并使用该包？

解决方案 #

由于intel_extension_for_pytorch暂未提供conda安装源，直接在激活的Anaconda环境中使用pip安装即可，步骤如下：

• 激活目标Anaconda环境
  打开终端，执行命令（替换your_env_name为你的环境名称）：

  conda activate your_env_name
  

• 确认PyTorch版本匹配
  该扩展包需要与环境中已安装的PyTorch版本严格对应，先用以下命令查看当前PyTorch版本：

  pip list | findstr torch  # Windows系统
  # Linux/Mac系统执行：pip list | grep torch
  

  若环境未安装PyTorch，先安装对应版本的CPU/Intel GPU兼容版，例如：

  pip install torch torchvision torchaudio
  

• 使用pip安装intel_extension_for_pytorch
  根据你的PyTorch版本，执行匹配的pip安装命令。例如使用PyTorch 2.1.x时，命令为：

  pip install intel-extension-for-pytorch==2.1.10+xpu
  

  若不确定具体版本号，可尝试自动匹配的基础安装命令：

  pip install intel-extension-for-pytorch
  

• 验证安装并使用
  安装完成后，可在Python中导入验证并使用：

  import torch
  import intel_extension_for_pytorch as ipex
  
  # 针对Intel GPU：将模型和数据移至xpu设备
  model = model.to("xpu")
  data = data.to("xpu")
  
  # 针对CPU优化：用ipex优化模型和优化器
  model, optimizer = ipex.optimize(model, optimizer=optimizer)
  

内容的提问来源于stack exchange，提问作者allinmoderation