Cloneofsimo Lora Versions Save

Using Low-rank adaptation to quickly fine-tune diffusion models.

v0.1.7

1 year ago

v0.1.6

1 year ago

What's Changed

New Contributors

Full Changelog: https://github.com/cloneofsimo/lora/compare/v0.1.5...v0.1.6

I've worked with team @replicate to improve this repo capable of various productions! Thank you so much!

v0.1.5

1 year ago

What's Changed

New Contributors

Full Changelog: https://github.com/cloneofsimo/lora/compare/v0.1.4...v0.1.5

v0.1.4

1 year ago

What's Changed

Now output CLIP alignment scores during lora tuning step, mitigated center-crop by default error, saves lora with extra argument name.

Full Changelog: https://github.com/cloneofsimo/lora/compare/v0.1.3...v0.1.4

v0.1.3

1 year ago

What's Changed

New Contributors

Full Changelog: https://github.com/cloneofsimo/lora/compare/v0.1.2...v0.1.3

v0.1.2

1 year ago

What's Changed

New Contributors

Full Changelog: https://github.com/cloneofsimo/lora/compare/v0.1.1...v0.1.2

v0.1.1

1 year ago

What's Changed

Full Changelog: https://github.com/cloneofsimo/lora/compare/v0.1.0-alpha...v0.1.1

v0.1.0-alpha

1 year ago

Better Pivotal Tuning, CLIP evaluation, SVD distillation!

image

Many tricks / features went into this release :

  1. Extended Latent textual inversion
  2. Pivotal tuning with stage-learning rate schedule
  3. Norm-prior Bayesian learning
  4. Mask conditioned score estimation (+ in-pipeline face detector)
  5. Sub-1 denoising
  6. CLIP score (text alignment score, image alignment score) evaluater
  7. xformer support (not in lorpt CLI yet)
  8. wandb logging
  9. Better Caption dataset
  10. Distilling fully trained model into LoRA : SVD distillation CLI

What's Changed

New Contributors

Full Changelog: https://github.com/cloneofsimo/lora/commits/v0.1.0-alpha