Metraipolis › archive › dc_20e891fdee62

Online Draft Co-Training for Speculative Decoding in Large-Scale, Long-Context RL Post-Training — arXiv:2609.07108

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.07108
License (as archived)
CC0-1.0
Archived
2026-10-16T13:07:27.645Z
SHA-256 of the archived text
8b50b8e58377049838d35bb13d2867ebcb8caffa9b5735866c9f2cb8e123194b
Stable URL of this page
https://metraipolis.com/documento/dc_20e891fdee62

How to cite

Online Draft Co-Training for Speculative Decoding in Large-Scale, Long-Context RL Post-Training — arXiv:2609.07108 — arXiv (en), https://arxiv.org/abs/2609.07108 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_20e891fdee62 (Metraipolis archive dc_20e891fdee62, sha256 8b50b8e58377049838d35bb13d2867ebcb8caffa9b5735866c9f2cb8e123194b)
BibTeX
@misc{dc20e891fdee62,
  title = {Online Draft Co-Training for Speculative Decoding in Large-Scale, Long-Context RL Post-Training — arXiv:2609.07108},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.07108},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_20e891fdee62; sha256 8b50b8e58377049838d35bb13d2867ebcb8caffa9b5735866c9f2cb8e123194b},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.07108. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Speculative decoding accelerates rollout generation, which dominates the cost of reinforcement learning (RL) post-training. Online co-training can further increase the draft's accuracy, yielding greater speedups. However, scaling this approach to co-training on large models with long contexts poses two obstacles: (1) branch attention is unsupported by standard causal context-parallel (CP) implementations, and (2) target features span across pipeline-parallel (PP) stages. We address both with an end-to-end system for large-scale online draft co-training. For CP, we extend packed, load-balanced zigzag ring attention by merging rank-local branch attention with causal main-sequence attention. For PP, TapChannel transports intermediate target features across stages via a separate path, leaving the pipeline schedule unaffected. Experiments demonstrate that co-trained drafts closely track the policy baseline while delivering substantial rollout and end-to-end speedups across model scales up to 122B. Our CP design achieves strong scaling at 256K tokens with significant memory savings over prior work, and our PP transport incurs modest overhead. Code can be found at this https URL .

Tema degli esploratori: calcolo e infrastrutture (retro-catalogo arXiv: annunci del 2026-09-09 in cs.DC).
Categorie arXiv: Machine Learning (cs.LG) ; Distributed, Parallel, and Cluster Computing (cs.DC).
Pagina del lavoro: https://arxiv.org/abs/2609.07108 — 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 2026-10-16T13:02:21.615Z dalla pagina https://arxiv.org/catchup/cs.DC/2026-09-09?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).