Mvsepless Space not working?
This space it's not working, but what?
Hey, so inference runs via CLI, which is great because it frees up RAM after the inference is done. But we ran into a major issue with how ZeroGPU works.
Turns out, ZeroGPU only works in the main process. If you try to run the splitting script via CLI from inside the Web-UI, the script just can’t access ZeroGPU — even if you explicitly set the device to cuda:0. It ends up running the split on the CPU instead.
That means your ZeroGPU quota could end up getting wasted.
So, hang tight for a space update. In the meantime, if you need to separate audio, you can do it on Google Colab with a T4 GPU. Here’s the notebook link: https://colab.research.google.com/github/noblebarkrr/mvsepless/blob/epsilon/MVSepLess_Epsilon_Colab_v2.ipynb