search

FangyunWei / SLRT

236 stars 46 forks source link

Minimal requirements #36

Closed sobek1886 closed 3 months ago

sobek1886 commented 10 months ago

Hi! I'm trying to run TwoStreamNetwork on my Mac machine.

Due to dependency issues, I had to remove platform-specific package build info, some packages which do not have osx builds in conda (libgomp, libstdcxx-ng, libgcc-ng, _openmp_mutex) as well as the specific version numbers from the environment.yml file. Without those measures I was not successful in creating the conda environment.

I anticipate that I will run into some issues later because of that and I wonder, what are the minimal requirements? Are the packages that I deleted from the .yml file necessary to run the project? If so, could you suggest some alternatives for OS X? For which packages is a specific version actually required?

2000ZRL commented 9 months ago

Honestly, I don't think TwoStream Network can be run on a Mac machine. We train and test it on GPU clusters. We use 8x32GB V100 for training, while a single 16GB GPU may handle the inference process.

sobek1886 commented 9 months ago

Hi, thanks for your answer. Now I understand that trying to run your suggested evaluation might have been too much for my Mac. I would still love to try using the pretrained models for inference though and in general I think adding a requirements.txt with minimal dependencies would be useful to make running inference easier on different systems and workflows (e.g. not using Conda)

2000ZRL commented 9 months ago

Thanks for your suggestions. However, the current version relies on the packages list in the requirements.txt.

Best wishes, Ronglai Zuo

发件人: pSobecki @.> 发送时间: Tuesday, December 5, 2023 5:55:49 PM 收件人: FangyunWei/SLRT @.> 抄送: Ronglai Zuo @.>; Comment @.> 主题: Re: [FangyunWei/SLRT] Minimal requirements (Issue #36)

Hi, thanks for your answer. Now I understand that trying to run your suggested evaluation might have been too much for my Mac. I would still love to try using the pretrained models for inference though and in general I think adding a requirements.txt with minimal dependencies would be useful to make running inference easier on different systems and workflows (e.g. not using Conda)

― Reply to this email directly, view it on GitHubhttps://github.com/FangyunWei/SLRT/issues/36#issuecomment-1840413347, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AKLOGIG2US3XFRASYQWJJSTYH3VSLAVCNFSM6AAAAAA64TUMZSVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQNBQGQYTGMZUG4. You are receiving this because you commented.Message ID: @.***>