Streamlit Cloud部署应用时安装torch_sparse遇torch模块缺失错误求助

问题描述 #

部署应用时，torch_sparse==0.6.16安装报错，错误信息显示ModuleNotFoundError: No module named 'torch'。尽管requirements.txt中已包含torch==1.8.1，且日志显示pip先下载了torch，但在生成torch_sparse的metadata时，系统仍找不到torch模块。

错误日志 #

Collecting torch==1.8.1
  Downloading torch-1.8.1-cp39-cp39-manylinux1_x86_64.whl (804.1 MB)
     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 804.1/804.1 MB 147.0 MB/s eta 0:00:00[2023-02-23 12:32:59.564581] 

Collecting matplotlib==3.5.2
  Downloading matplotlib-3.5.2-cp39-cp39-manylinux_2_5_x86_64.manylinux1_x86_64.whl (11.2 MB)
     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 11.2/11.2 MB 90.2 MB/s eta 0:00:00

Collecting translate==3.6.1
Collecting torch_sparse==0.6.16
  Downloading torch_sparse-0.6.16.tar.gz (208 kB)
     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 208.2/208.2 KB 285.3 MB/s eta 0:00:00[2023-02-23 03:52:24.740968] 
  Preparing metadata (setup.py): started
  Preparing metadata (setup.py): finished with status 'error'
  error: subprocess-exited-with-error
  × python setup.py egg_info did not run successfully.
  │ exit code: 1
  ╰─> [6 lines of output]
      Traceback (most recent call last):
        File "<string>", line 2, in <module>
        File "<pip-setuptools-caller>", line 34, in <module>
        File "/tmp/pip-install-6ook3r5m/torch-sparse_a4c09b3da42046a5a246b2ccd6433d71/setup.py", line 8, in <module>
          import torch
      ModuleNotFoundError: No module named 'torch'
      [end of output]

  note: This error originates from a subprocess, and is likely not a problem with pip.
error: metadata-generation-failed
× Encountered error while generating package metadata.
╰─> See above for output.
note: This is an issue with the package mentioned above, not pip.
hint: See above for details.

当前requirements.txt内容 #

torch==1.8.1
matplotlib==3.5.2
translate==3.6.1
numpy==1.21.6
pandas==1.3.5
streamlit==1.15.1
plotly==5.11.0
scipy==1.7.3
networkx==2.6.3
rdkit==2022.9.2
requests==2.28.1
Pillow==9.4.0
scikit_learn==1.2.1
torch_sparse==0.6.16
torch_geometric==2.1.0.post1

解决方案 #

方案1：使用预编译wheel包（推荐） #

PyTorch Geometric提供了对应不同torch版本的预编译wheel，避免从源码编译时依赖torch。修改requirements.txt，在开头添加对应torch版本的索引源：

--extra-index-url https://data.pyg.org/whl/torch-1.8.0+cu111.html
torch==1.8.1
matplotlib==3.5.2
translate==3.6.1
numpy==1.21.6
pandas==1.3.5
streamlit==1.15.1
plotly==5.11.0
scipy==1.7.3
networkx==2.6.3
rdkit==2022.9.2
requests==2.28.1
Pillow==9.4.0
scikit_learn==1.2.1
torch_sparse==0.6.16
torch_geometric==2.1.0.post1

注意：索引地址需匹配你的torch版本，比如torch-1.8.0+cu111对应torch 1.8.x版本+CUDA 11.1，若使用CPU版本则替换为torch-1.8.0+cpu。

方案2：使用安装脚本强制安装顺序 #

创建setup.sh文件，手动指定安装顺序，确保torch完全安装后再安装依赖它的包：

pip install torch==1.8.1
pip install torch_sparse==0.6.16 torch_geometric==2.1.0.post1
pip install -r requirements.txt

然后创建.streamlit/config.toml文件，禁用自动依赖安装，让Streamlit执行自定义脚本：

[server]
install_dependencies = false

将这两个文件上传到GitHub仓库，Streamlit Cloud会优先执行setup.sh完成安装。

方案3：禁用构建隔离 #

在requirements.txt中给torch_sparse添加--no-build-isolation参数，避免pip在隔离环境中编译该包：

torch==1.8.1
torch_sparse==0.6.16 --no-build-isolation
torch_geometric==2.1.0.post1
# 其他依赖...

内容的提问来源于stack exchange，提问作者cuger