AstolfoXL (WIP)

Probably the first (and the only) individual Full Finetuning with multi-GPU.

LoKR works, but no thanks. 我不做人了，早苗！

Discord: "Good luck".

Specification

• Base model: AstolfoMix-XL, version 255c

• Tech report: ch06

• Training metrics (tensorboard): HF

• Dataset (images > latents): danbooru2024, e621_2024

• Dataset (tags + captions): meta_lat.json

• 1 step = 16 images, 4x RTX 3090 24G.

• 779k steps for 1EP, 8.0 + 4.6 = 12.6M images

• Tag + NLP caption with A1111 token trick

• Trainer codes: The PR won't be merged

• Train parameter: adamW8bit, UNET 1.5e-6, TE 1.2e-5, BS4 (4 GPU) grad accu 4, 71% UNET (Speed + must underfit)

• 75-100+ days for 1EP. Train 1 EP only. Save per 10k steps.

• Core concept: Unsupervised learning

• Expectation: MID (100% no filter no quality tag)

How to use

• Train LoRA / merge on top of this model. Compatability should still close to 215c. Realistic human content is still supported. "Trust me bro".

• Artist tags may not work, but I did trained. Just dump your "NAI" prompts here.

• Use TIPO to expand tag based prompts with NLP.

• Short tags will suffer from background latent noise. Tags can be observed from E621 or danbooru.

• Any observation before 1EP should not being referenced for justification. All images are just seen once.

Full docuementation will be published, which is as long as the AstolfoMix series.