search

YuanGongND / ssast

Code for the AAAI 2022 paper "SSAST: Self-Supervised Audio Spectrogram Transformer".
BSD 3-Clause "New" or "Revised" License
365 stars 61 forks source link

python and cuda version #23

Open nathanchenseanwalter opened 1 year ago

nathanchenseanwalter commented 1 year ago

what version python3 and cuda was this used on

YuanGongND commented 1 year ago

I think it was run with Python 3.7, cuda 10.1, and cudnn 7.6.

Is there a reason makes higher version a problem?

nathanchenseanwalter commented 1 year ago

On my end (after some trial and error), versions that are lower than a specific set of 3.7 or higher than a specific set of 3.8 can't install the requirements. I think the biggest issue I'm running into is with scikit-learn.

The cuda issue I'm running into is because one of my projects runs locally on WSL which only supports CUDA after 11, so torch would run into issues communicating with the cuda kernel if I do end up installing as previous.

YuanGongND commented 1 year ago

I see thanks for reporting this.

For Python, we use 3.7.4, the experiment was done on Ubuntu machines. Isn't scikit-learn a commonly used package that supports major Python versions?

I guess on Ubuntu, CUDA 11.0 would be fine. We don't have tricky thing that is only runnable on lower versions. What is WSL?

-Yuan