RAPIDS pip安装问题

huangapple go评论109阅读模式
英文:

RAPIDS pip installation issue

问题

我一直在尝试在我的Docker环境中安装RAPIDS,最初进展顺利。然而,在过去的一两周里,我一直遇到一个错误。

问题似乎是pip试图从默认的PyPi注册表中获取,其中遇到了一个占位符项目。我不确定是谁放置它在那里,或者为什么,因为它似乎没有实际用途。

我尝试将--index-url明确设置为pypi.nvidia.com,但这种方法也不可行,因为RAPIDS软件包的依赖似乎托管在默认的PyPi上。

英文:

I've been trying to install RAPIDS in my Docker environment, which initially went smoothly. However, over the past one or two weeks, I've been encountering an error.

The issue seems to be that pip is attempting to fetch from the default PyPi registry, where it encounters a placeholder project. I'm unsure who placed it there or why, as it appears to serve no practical purpose.

 => ERROR [12/19] RUN pip3 install cudf-cu11 cuml-cu11 cugraph-cu11 cucim --extra-index-url=https://pypi.nvidia.com                                             2.1s 
------                                                                                                                                                               
 > [12/19] RUN pip3 install cudf-cu11 cuml-cu11 cugraph-cu11 cucim --extra-index-url=https://pypi.nvidia.com:                                                        
#0 1.038 Looking in indexes: https://pypi.org/simple, https://pypi.nvidia.com                                                                                        
#0 1.466 Collecting cudf-cu11                                                                                                                                        
#0 1.542   Downloading cudf-cu11-23.6.0.tar.gz (6.8 kB)                                                                                                              
#0 1.567   Preparing metadata (setup.py): started
#0 1.972   Preparing metadata (setup.py): finished with status 'error'
#0 1.980   error: subprocess-exited-with-error
#0 1.980   
#0 1.980   × python setup.py egg_info did not run successfully.
#0 1.980   │ exit code: 1
#0 1.980   ╰─> [16 lines of output]
#0 1.980       Traceback (most recent call last):
#0 1.980         File "<string>", line 2, in <module>
#0 1.980         File "<pip-setuptools-caller>", line 34, in <module>
#0 1.980         File "/tmp/pip-install-8463q674/cudf-cu11_9d3e1a792dae4026962cdff29926ce8d/setup.py", line 137, in <module>
#0 1.980           raise RuntimeError(open("ERROR.txt", "r").read())
#0 1.980       RuntimeError:
#0 1.980       ###########################################################################################
#0 1.980       The package you are trying to install is only a placeholder project on PyPI.org repository.
#0 1.980       This package is hosted on NVIDIA Python Package Index.
#0 1.980       
#0 1.980       This package can be installed as:
#0 1.980       ```
#0 1.980       $ pip install --extra-index-url https://pypi.nvidia.com cudf-cu11
#0 1.980       ```
#0 1.980       ###########################################################################################
#0 1.980       
#0 1.980       [end of output]
#0 1.980   
#0 1.980   note: This error originates from a subprocess, and is likely not a problem with pip.
#0 1.983 error: metadata-generation-failed
#0 1.983 
#0 1.983 × Encountered error while generating package metadata.
#0 1.983 ╰─> See above for output.
#0 1.983 
#0 1.983 note: This is an issue with the package mentioned above, not pip.
#0 1.983 hint: See above for details.

I attempted to explicitly set the --index-url to pypi.nvidia.com, but this approach wasn't feasible either, as the dependencies for the RAPIDS packages appear to be hosted on the default PyPi.

答案1

得分: 1

问题已经在 cudf 存储库中报告,并且似乎从版本 23.06 开始存在。

我将尝试安装版本 23.04 的建议解决方法,如果这暂时解决了问题,我将报告回来。

原始问题:
https://github.com/rapidsai/cudf/issues/13642

反馈:
使用严格的版本号 23.04 完美运行。我只是用带版本号的 pip 安装替换了无版本号的一个:

# 之前
pip3 install cudf-cu11 cuml-cu11 cugraph-cu11 cucim --extra-index-url=https://pypi.nvidia.com

# 之后
pip3 install cudf-cu11==23.04 cuml-cu11==23.04 cugraph-cu11==23.04 cucim --extra-index-url=https://pypi.nvidia.com

对于在 Ubuntu 20.04 容器中构建时遇到此问题的其他人需要注意一点: 在安装 rapids 之前,您需要添加 RUN apt remove python3-psutil,以允许 pip 安装正确版本的 psutil。

英文:

The issue has already been reported in the cudf repository and it seems to have existed since version 23.06.

I will try the suggested workaround of installing version 23.04 and will report back if this temporarily resolves the problem.

Original Issue:

https://github.com/rapidsai/cudf/issues/13642

Feedback:

Using strict versioning to 23.04 worked perfectly.
I simply replaced the unversioned pip install with a versioned one:

# Before
pip3 install cudf-cu11 cuml-cu11 cugraph-cu11 cucim --extra-index-url=https://pypi.nvidia.com

# After
pip3 install cudf-cu11==23.04 cuml-cu11==23.04 cugraph-cu11==23.04 cucim --extra-index-url=https://pypi.nvidia.com

One thing to note for anyone else encountering this issue while building in an Ubuntu 20.04 container: you need to add RUN apt remove python3-psutil before installing rapids, to allow pip to install psutil in the correct version.

huangapple
  • 本文由 发表于 2023年6月29日 17:27:37
  • 转载请务必保留本文链接:https://go.coder-hub.com/76579783.html
匿名

发表评论

匿名网友

:?: :razz: :sad: :evil: :!: :smile: :oops: :grin: :eek: :shock: :???: :cool: :lol: :mad: :twisted: :roll: :wink: :idea: :arrow: :neutral: :cry: :mrgreen:

确定