Simple GUI for finetrainers (but in no way affiliated with it) made using gradio, inspired by bmalthais gui for kohya-ss scripts.
The goal is to be able to reproduce and analyze previous runs by saving meta data.
Doesn't contain cogvideox-factory. You must download it and get it running separately.
I try to keep main stable, but if it fails, step back one version and try that.
v0.10.0: fp8 training
v0.9.1: Add 'resume_from_checkpoint' setting
v0.9.0: Add missing precompute_condition settings
v0.8.0: Configuration validator. Fail early.
Tabs for training and prepare dataset (legacy)
I'm migrating to the new finetrainers structure. Currently tested with LTX-Video. Keeping legacy CogVideoX trainer in the "Legacy" tab.
Usage:
Have CogvideoX-Factory finetrainers
Clone repo
Install streamlit pip install gradio
Run with python app.py
Config path to cogvideox-factory in app
Configure your runtime settings
Press run
Cross fingers and hope it works
Early days: May not work as expected.
The editor is pretty dumb at the moment, and formatting is expected to be the same as in the .sh files, ie spaces between buckets, comma between gpu_ids.
Tip:
If I've missed any configuration options you want, you can add them to configs/config_template.yaml and they will be picked up by the app.