Problem running this on multiple GPUs

[Ofplarsen]

Trouble when running this code on a cluster of GPUs

Lightning-AI/torchmetrics#2477

[danelee2601]

The current code doesn't support the multiple gpu yet. But I'm planning to do update it.

[danelee2601]

Thanks for the feedback.

[Ofplarsen]

No problem, I'll manage with a single. Nice repo btw!

[Frndznz]

What would be the steps required to do this? @danelee2601 I have quite a large dataset and need to train on multiple GPUs, and I've managed to train on multiple GPUs but only using the deepseed strategy for PyTorch Lightning. Have you made any progress on this?

[danelee2601]

Hi @Frndznz. This repo hasn't bee developed to train on a very large dataset that can't be fit into CPU memory, as it was developed for a research purpose. But I'm sure you can modify the data pipeline to enable it. For instance, save a csv file that contains a list of all the available data file names and import the files for every iteration.