Metraipolis › archive › dc_7bc14de36381
Abra: Scaling Diffusion Image Training — arXiv:2608.17286
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2608.17286
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-18T14:48:33.434Z
- SHA-256 of the archived text
b6af01bdce2d7261b84b883cf48c381b85b300c01959caf8ba9ad5e74064a050- Stable URL of this page
https://metraipolis.com/documento/dc_7bc14de36381
How to cite
Abra: Scaling Diffusion Image Training — arXiv:2608.17286 — arXiv (en), https://arxiv.org/abs/2608.17286 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_7bc14de36381 (Metraipolis archive dc_7bc14de36381, sha256 b6af01bdce2d7261b84b883cf48c381b85b300c01959caf8ba9ad5e74064a050)
BibTeX
@misc{dc7bc14de36381,
title = {Abra: Scaling Diffusion Image Training — arXiv:2608.17286},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2608.17286},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_7bc14de36381; sha256 b6af01bdce2d7261b84b883cf48c381b85b300c01959caf8ba9ad5e74064a050},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.17286. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Compute-optimal scaling laws guide the training of frontier language models yet remain largely unexplored for visual generation. We present a systematic scaling law study for text-to-image diffusion models using Abra, a controlled family of flow-matching transformers trained across three orders of magnitude worth of compute ($10^{19}$ to $10^{22}$ FLOPs), reaching significantly larger compute budgets than previous works. We demonstrate that diffusion models scale just as predictably as language models but require far more data to train optimally: compute optimality occurs at approximately $200$ image tokens per parameter, ten times the Chinchilla compute-optimal prescription for LLMs. We show that unlike language models, diffusion models are robust to overtraining and that practitioners should err on the side of more data rather than a larger model. Finally, we show that this predictability extends beyond training loss to generative quality metrics, optimal CFG settings, representation quality, and even the shape of the training curves, which collapse onto a universal form.
Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-08-19 in cs.LG).
Categorie arXiv: Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2608.17286 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-18T14:47:59.391Z dalla pagina https://arxiv.org/catchup/cs.LG/2026-08-19?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).