Mac M1 Monterey环境下Poetry安装Torch报错：找不到nvidia-cudnn-cu11安装候选

问题描述 #

运行环境：Mac M1 Monterey，干净环境，Python 3.8.13，Poetry 1.2.2（按官方流程安装）

执行poetry add torch安装Torch时触发RuntimeError，提示无法找到nvidia相关依赖包的安装候选，错误日志如下：

✗ python --version    
Python 3.8.13

✗ poetry --version
Poetry (version 1.2.2)

✗ poetry add torch
Using version ^1.13.0 for torch

Updating dependencies
Resolving dependencies... (0.1s)

Package operations: 5 installs, 0 updates, 0 removals

  • Installing nvidia-cublas-cu11 (11.10.3.66): Pending...
  • Installing nvidia-cuda-nvrtc-cu11 (11.7.99): Failed

  RuntimeError

  • Installing nvidia-cuda-nvrtc-cu11 (11.7.99): Failed

  RuntimeError

  • Installing nvidia-cublas-cu11 (11.10.3.66): Failed
  • Installing nvidia-cuda-nvrtc-cu11 (11.7.99): Failed


  • Installing nvidia-cudnn-cu11 (8.5.0.96): Failed

  RuntimeError

  Unable to find installation candidates for nvidia-cudnn-cu11 (8.5.0.96)

  at ~/Library/Application Support/pypoetry/venv/lib/python3.8/site-packages/poetry/installation/chooser.py:103 in choose_for
       99│ 
      100│             links.append(link)
      101│ 
      102│         if not links:
    → 103│             raise RuntimeError(f"Unable to find installation candidates for {package}")
      104│ 
      105│         # Get the best link
      106│         chosen = max(links, key=lambda link: self._sort_key(package, link))
      107│ 

但使用pip install torch可正常完成安装：

✗ pip install torch        
Collecting torch
  Downloading torch-1.13.0-cp38-none-macosx_11_0_arm64.whl (55.3 MB)
     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 55.3/55.3 MB 3.0 MB/s eta 0:00:00
Requirement already satisfied: typing-extensions in /Users/vincent/Library/Caches/pypoetry/virtualenvs/recommend-drawify-iwMS_JJ1-py3.8/lib/python3.8/site-packages (from torch) (4.4.0)
Installing collected packages: torch
Successfully installed torch-1.13.0

解决方法 #

问题本质是Poetry默认尝试拉取适配NVIDIA CUDA的Torch包，但Mac M1为ARM架构且无NVIDIA GPU，不需要这些CUDA依赖。可通过以下方式解决：

• 指定平台安装
  直接在命令中指定Mac M1适配的版本：

poetry add torch@1.13.0 --platform macos_arm64

• 配置pyproject.toml
  在项目的pyproject.toml中明确Torch的平台约束，确保Poetry优先选择适配包：

[tool.poetry.dependencies]
python = "^3.8"
torch = { version = "^1.13.0", platform = "macos_arm64" }

• 借助pip过渡
  若上述方法无效，先通过pip install torch完成安装，再执行poetry add torch让Poetry识别已安装包，或直接将Torch版本信息写入pyproject.toml后执行poetry install。

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