Metraipolis › archive › dc_e0d4ca4dfd62
Recovering Off-Policy Supervision for Speculative Decoding — arXiv:2609.38795
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.38795
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-07T13:13:30.315Z
- SHA-256 of the archived text
b8987944f21064c423f1c3f00149e3b5cf9c2cba4d1e79c933491ca3e3b762bf- Stable URL of this page
https://metraipolis.com/documento/dc_e0d4ca4dfd62
How to cite
Recovering Off-Policy Supervision for Speculative Decoding — arXiv:2609.38795 — arXiv (en), https://arxiv.org/abs/2609.38795 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_e0d4ca4dfd62 (Metraipolis archive dc_e0d4ca4dfd62, sha256 b8987944f21064c423f1c3f00149e3b5cf9c2cba4d1e79c933491ca3e3b762bf)
BibTeX
@misc{dce0d4ca4dfd62,
title = {Recovering Off-Policy Supervision for Speculative Decoding — arXiv:2609.38795},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.38795},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_e0d4ca4dfd62; sha256 b8987944f21064c423f1c3f00149e3b5cf9c2cba4d1e79c933491ca3e3b762bf},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.38795. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Block drafters for speculative decoding are commonly trained on corpora written by external models, where a single off-policy token invalidates supervision for all subsequent slots in a block. Existing approaches discard these divergent slots, resulting in severe supervision loss. To resolve this problem while preserving the training corpus, we propose a rollout-based training framework that recovers full supervision through two complementary components. The first component, Anchor-Label Relabelling (ALR), replaces corpus labels with distributions from greedy target rollouts, restoring valid supervision across all predicted slots. The second component, In-Rollout Anchors (IRA), places draft blocks directly inside these rollouts to expose the drafter to target-generated context, reusing precomputed rollout features at no additional target cost. Across fixed vision-language and text corpora, our framework increases greedy accepted length by up to 36.5% over DFlash and consistently outperforms erasing baselines. Notably, a single epoch of our method surpasses the best erase schedules. After three epochs, it matches the acceptance length of training on target-regenerated responses. These results show that our framework provides an effective and compute-efficient approach for training speculative drafters on fixed corpora without modifying the original text. Code is available at this https URL . Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-10-01 in cs.CL). Categorie arXiv: Computation and Language (cs.CL) ; Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG). Pagina del lavoro: https://arxiv.org/abs/2609.38795 — 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 2026-10-07T12:44:51.755Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-10-01?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).