Benjamin Bossan

Thanks for the further investigation @Haakooto. I have a tentative solution in #2011, it would be great if you could test it. I consider this WIP as I'm uncertain if...

Thanks @Haakooto for testing, this is good to know. Let's focus on OLoRA for now and deal with PiSSA later. I discovered a bug in the PR as it didn't...

@44670 The PR is still not merged, but you could install from the `https://github.com/BenjaminBossan/peft/tree/fix-olora-bnb` branch if you want to test it right now.

@hoang1645 Could you please provide a reproducer and the full error message?

The accelerate version you use is too old, the argument was introduced in v0.26. If you upgrade to the latest version, v0.28, the way to pass the argument is like...

Thanks for the report. We do have an [example of using pytorch geometric](https://github.com/skorch-dev/skorch/blob/master/notebooks/CORA-geometric.ipynb) with a section on data handling, not sure if this can be applied to your use case....

Hmm, I wonder if it's really related or not, because the latest finding in that thread seems to be that memory has stabilized but latency goes up. Here we still...

> the other thread is related to GPU memory while I'm facing this issue in CPU memory (RAM) The code you showed is, however, using the GPU, or am I...

Thanks for providing more details. Just to be completely clear, so that I can attempt to replicate: 1. The model is on GPU but you observe a memory leak on...

I could reproduce these results. However, I don't think they're related to diffusers or PEFT. Instead, they seem to be related to safetensors. To demonstrate, the snippet below shows the...