Replies: 2 comments 1 reply
-
Hi, |
Beta Was this translation helpful? Give feedback.
1 reply
-
After training with networks.oft, I couldn't successfully use the resulting file as a LoRA in ComfyUI. Any tips on how I could fix that? |
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
How to Train GLoRA and OFT using LyCORIS
This is a high-level practical guide on how to train these new networks using LyCORIS and kohya-ss' sd-scripts.
Both networks have improved performance over LoRA in many tasks. Refer to the individual papers for metrics.
To train GLoRA or OFT with LyCORIS, ensure your version of LyCORIS is updated to >= v2.0.0.
1. GLoRA with LyCORIS
Training command line arguments:
Relevant links:
Paper: https://arxiv.org/abs/2306.07967
Official implementation: https://github.com/Arnav0400/ViT-Slim/tree/master/GLoRA
Summary/discussion of paper: https://arize.com/blog/generalized-lora-glora-paper-reading/
LyCORIS implementation: https://github.com/KohakuBlueleaf/LyCORIS/blob/main/lycoris/modules/glora.py
2. OFT
Supports SDXL only!
Training OFT/COFT with LyCORIS
Training command line arguments:
network_dim: Controls the size of the orthogonal block. Values > 32 approach full fine-tuning (out of memory).
network_alpha: Ignored in training / inference
constrain (ϵ): Controls how "flexible" the model is.
rescaled (Appendix C in the paper): Trains a lightweight additional parameter to improve visual quality of images and to correct color distortion that may result from the re-orientation of the neurons. Probably best set to True.
COFT with kohya-ss networks.oft
Training command line arguments:
Relevant links:
Paper: https://arxiv.org/abs/2306.07280
Official implementation: https://github.com/Zeju1997/oft
Project page: https://oft.wyliu.com/
LyCORIS implementation: https://github.com/KohakuBlueleaf/LyCORIS/blob/main/lycoris/modules/diag_oft.py
kohya-ss implementation: https://github.com/kohya-ss/sd-scripts/blob/main/networks/oft.py
Beta Was this translation helpful? Give feedback.
All reactions