Metraipolis › archive › dc_520c0cb5cdb3
Allspark: Weak to Strong Transfer via Alternating Chain of Thought — arXiv:2609.32913
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.32913
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-07T21:00:46.694Z
- SHA-256 of the archived text
319509ce5ede21c909a20868afb4c1f4a7615d61ad593828deb09a60a6037c39- Stable URL of this page
https://metraipolis.com/documento/dc_520c0cb5cdb3
How to cite
Allspark: Weak to Strong Transfer via Alternating Chain of Thought — arXiv:2609.32913 — arXiv (en), https://arxiv.org/abs/2609.32913 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_520c0cb5cdb3 (Metraipolis archive dc_520c0cb5cdb3, sha256 319509ce5ede21c909a20868afb4c1f4a7615d61ad593828deb09a60a6037c39)
BibTeX
@misc{dc520c0cb5cdb3,
title = {Allspark: Weak to Strong Transfer via Alternating Chain of Thought — arXiv:2609.32913},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.32913},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_520c0cb5cdb3; sha256 319509ce5ede21c909a20868afb4c1f4a7615d61ad593828deb09a60a6037c39},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.32913. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Recent progress in frontier models has renewed interest in large-scale reinforcement learning (RL), but the cost of generating large-model rollouts makes even testing RL recipes expensive. We ask whether reasoning improvements learned by a small, weak model can benefit a larger, stronger model without using the strong model's rollouts during training. We introduce Allspark, a training and inference framework for weak-to-strong transfer through alternating chains of thought. A weak teacher is trained alongside a frozen copy of the same model; the two alternate reasoning segments, and the frozen model produces the final answer. At inference time, a stronger student replaces the frozen training partner, while both models remain fixed. Because they communicate through text, the teacher can steer students from different model families and with different tokenizers. We study Allspark at two scales: controlled Qwen experiments across math and reasoning, and larger-scale Inkling experiments on ARC-AGI-2. The Inkling experiments show accuracy gains in within-family and cross-family settings, including transfer to Kimi and Nemotron, with benefits that vary across inference settings. These findings motivate reusing a trained weak teacher across strong students and examining the resulting accuracy--token tradeoff. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-29 in cs.AI). Categorie arXiv: Machine Learning (cs.LG) ; Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2609.32913 — 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 2026-10-07T20:29:02.500Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-29?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).