search

Captain-FLAM / KaraFan

The BEST music separation model with help of A.I. ... to my ears ! 👂👂
MIT License
129 stars 12 forks source link

Code obsolescence #11

Open BauTsAnD opened 8 months ago

BauTsAnD commented 8 months ago

Code is becoming obsolete in new versions of Python and Google colab The problems are slowness and incompatibility in Google colab with the GPU

deton24 commented 8 months ago

Make a new cell with: !apt-get install cuda-11-8 added after installation cell, and separation time is fixed. But executing it will take 8 minutes.

For some reason, faster !pip3 install ort-nightly-gpu --index-url=https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/ort-cuda-12-nightly/pypi/simple/ or !python -m pip install ort-nightly-gpu --index-url=https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/ort-cuda-12-nightly/pypi/simple/ doesn't fix the issue like in other MDX-Net or MDX23C Colabs. The GUI behaves, like it wasn't using it at all.

deton24 commented 8 months ago

No need to wait 8 minutes anymore. They released official non-nightly onnxruntime-cuda-12 package, and it's compatible with KaraFan (contrary to the nightly package).

Use this Colab with the fix: https://colab.research.google.com/github/Eddycrack864/KaraFan/blob/master/KaraFan_Improved_Version.ipynb (with new models)

Or: Click on the first hidden line, find the last line of the code, go to the end of it (set cursor on the left of "(1)" ), press enter, and in the next line with the space added automatically on the left before the empty line when you pressed enter, paste: !pip install onnxruntime-gpu --extra-index-url https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/onnxruntime-cuda-12/pypi/simple/