How to install lightning-flash for specific cuda version

I want to install an older version of lightning-flash, which would run with cuda 10.1. I know there is an option for torch described at But I couldn’t find instructions anywhere for lightning-flash.
Thank you very much!

PS: I hope that this forum is also for lightning and flash. If it’s not the case, I apologize.

I’m not deeply familiar with flash, but based on the repository I don’t think there is any CUDA dependency, as only Python seems to be used (I also couldn’t find any .cu files there).

Thank you for your response. The problem is that I cannot train on GPU because when installing flash, it installs the newest version of the torch, which is not compatible with old cuda => it’s not possible to train on GPU. So to clarify the question: Is it possible to install an older version of lightning-flash with a specific version of the torch?

I don’t know which dependencies lightning-flash uses, but based on your description it seems a CPU-only PyTorch version is installed. Maybe pip install --no-dependencies ... would help to skip replacing PyTorch with the CPU binary.