TODO list

[p1atdev]

• Training efficiency
  • Latent & Text encoder output caching #2
• More models
  • Flux, Flex #3
  • SD3.5
  • Change repository name
• Logging
  • Better loss visualization (https://github.com/spacepxl/demystifying-sd-finetuning)
  • Validation loss
• Training methods
  • Shortcut Model https://arxiv.org/abs/2410.12557
    • currently not working well 🤔
• More peft features
  • Load pretrained peft weights
  • Weight initialization options
  • Assistant adapter (like ai-toolkit)