Kohaku is a cute dragon girl.

LyCORIS 2.0.0

By kblueleaf on Dec 24, 2023
Image post

LyCORIS 2.0.0

We are excited to announce that LyCORIS project has just been updated to version 2.0.0. This update brings a plethora of new features and improvements!

🎉HCP-Diffusion supports🎉

LyCORIS is now starting to support training/inference in HCP-Diffusion

  • Now LyCORIS support LoHa/LoKr/Diag-OFT algorithm in HCP-Diffusion
  • Add Pivotal tuning utilities
  • Add hcp convert utilities
  • Have no plan at this time to support full/lora and train_norms since HCP can do them natively

🌟New Implemented Algorithms🌟

🚀New features🚀

  • Standalone usage (For any pytorch module):
    • Can wrap any pytorch module which contains Linear/Conv2d/LayerNorm/GroupNorm modules
    • A project which utilize LyCORIS to finetune Phi-1.5: HakuPhi
    • minimal example:
from lycoris import create_lycoris, LycorisNetwork

    {"target_name": [".*attn.*"]}
lycoris_net = create_lycoris(

# after apply_to(), your_model() will run with LyCORIS net
lycoris_param = lycoris_net.parameters()
forward_with_lyco = your_model(x)
  • Merge scripts now support SDXL
  • Extract locon script now support SDXL

Fixed Bugs

  • Dropout have inversed rate. (dropout=a will performed as dropout=(1-a))
  • appy_max_norms will have divided by zero error
  • cannot resume correctly
© Copyright 2024 by Kohaku BlueLeaf.
All media on this site, including text, images, and videos, is licensed under CC BY-NC-ND 4.0