[FEATURE REQUEST] PEFT (Parameter-Efficient Fine-Tuning) Lora training for sdxl on t4 colab??

hollowstrawberry / kohya-colab

Accessible Google Colab notebooks for Stable Diffusion Lora training, based on the work of kohya-ss and Linaqruf

GNU General Public License v3.0

592 stars 83 forks source link

[FEATURE REQUEST] PEFT (Parameter-Efficient Fine-Tuning) Lora training for sdxl on t4 colab?? #202

Closed Bercraft closed 3 days ago

Bercraft commented 2 weeks ago

https://huggingface.co/docs/peft/v0.12.0/index https://github.com/huggingface/peft

If i understood correctly with this tech is possible to train sdxl Lora on low mem gpu, I do not know colab well otherwise i would have made a laptop myself, do you think you can make one with this tech or add to an existing one?

Link on Lora implementation from hugign face https://huggingface.co/docs/peft/main/en/developer_guides/lora

Thank you.

hollowstrawberry commented 2 weeks ago

Implementing something like this is sadly beyond the scope of this colab, as I don't think it's part of kohya 🤔

Bercraft commented 3 days ago

ok...thank you