Info, Guides and Lessons Learned - Nerogar/OneTrainer GitHub Wiki

Here you can find various pages for guides, info and lessons learned

Misc Info

For SD3 and Flux (gated models in general) training you need a read token from HuggingFace (In HF go in your settings and create a read token) and set it in as an environment variable named "HF_TOKEN". Other solutions exist like a Command line instruction (Discord discussion) or using a local copy (Discord discussion) of the HF folder.
Caption tokens limit: actually, most of them are limited to 77 tokens (75 without start and end tokens) because of clip. The only exceptions are: pixart (alpha/sigma): 120, sana: 300. May change in the future.
Common Mistakes Coming From Kohya If you're failing to get similar results that you could get in Kohya.
Validation explained by its author

OneTrainer March 2024 Guide (WIP) A more exhaustive guide on OneTrainer built as a checklist.
Cloud Training The OT native and supported way of running on Runpod or any remote Linux server to train.
Manually setup OneTrainer in Runpod Outdated since we now natively support it in the UI, but still an option.