lora-training

lora-training copied to clipboard

• [ ] ensure descriptions are useful/accurate. example: class_data mentions class_prompt and num_class_images that don't exist
• [ ] I've not seen other models expose device as an option... should we in lora-advanced-trainer?
• [ ] continue_inversion - unless we expose sending in existing tensor(s) - does this make sense?
• [ ] perform_inversion / train_text_encoder -- do both of these matter if we only return the last step_1000.safetensors?

Checked all the stuff you mentioned. train_text_encoder must be on in current implementation, but I'm going to update it in the future. #5