Metraipolis › archive › dc_e8fe63e26e01
arXiv 2610.28619: Adaptive Multi-Token Prediction: How far should we go?
- Source
- Metraipolis archive
- Original
- https://arxiv.org/abs/2610.28619
- License (as archived)
- CC-BY-4.0
- Archived
- 2026-10-21T19:08:32.246Z
- SHA-256 of the archived text
27ffdba759f77c6fba9948783e15b20c27061c0d6214440b7c6fda4bba83fdb6- Stable URL of this page
https://metraipolis.com/documento/dc_e8fe63e26e01
How to cite
arXiv 2610.28619: Adaptive Multi-Token Prediction: How far should we go? — Metraipolis archive, https://arxiv.org/abs/2610.28619 (CC-BY-4.0); archived copy: https://metraipolis.com/documento/dc_e8fe63e26e01 (Metraipolis archive dc_e8fe63e26e01, sha256 27ffdba759f77c6fba9948783e15b20c27061c0d6214440b7c6fda4bba83fdb6)
BibTeX
@misc{dce8fe63e26e01,
title = {arXiv 2610.28619: Adaptive Multi-Token Prediction: How far should we go?},
howpublished = {Metraipolis archive},
url = {https://arxiv.org/abs/2610.28619},
note = {License CC-BY-4.0; archived copy https://metraipolis.com/documento/dc_e8fe63e26e01; sha256 27ffdba759f77c6fba9948783e15b20c27061c0d6214440b7c6fda4bba83fdb6},
year = {2026}
}Archived text
Reproduced under CC-BY-4.0. Attribution: Metraipolis archive, https://arxiv.org/abs/2610.28619. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Multi-token prediction (MTP) can accelerate language-model decoding by drafting several tokens and verifying them in a single target-model pass. However, the best speculation horizon changes with context: longer drafts offer greater amortization when reliable, but can waste compute on suffixes discarded after a rejection. Unlike autoregressive drafting, where a stopping rule can decide token by token, MTP must fix the horizon before drafting, as all heads fire in the same forward pass. We therefore formulate horizon selection as an online control problem, using verification feedback to learn how far the draft can be trusted in each context, and introduce Adaptive Multi-Token Prediction (AMTP): a lightweight controller that requires no retraining and leaves the exact verifier unchanged. To guide horizon selection, we characterize the expected number of tokens committed at each horizon via acceptance survival, the probability that a draft prefix is fully accepted. Before each round, AMTP predicts how far the draft is likely to survive and picks the horizon with the highest expected tokens per second, learning freely from every verification pass. Theoretically, we show that any online learner's prediction error translates into throughput regret against the best horizon for each context, and in turn into a bound on wall-clock completion time. With online gradient descent, this regret is sublinear. Across five draft-target pairings built on Llama-3.1-8B-Instruct with Medusa heads, Meta's MTP models, and Qwen3-4B-Inst-2507-MTP, evaluated on AIME 2025 and GSM8K, AMTP decodes up to 4.1x faster than standard decoding and matches or outperforms the fixed window in 8 of 10 settings, with throughput gains of up to 10%. Articolo: «Adaptive Multi-Token Prediction: How far should we go?» — Anand, Emile; Parwani, Mohammed Rehan; Cao, Xinyuan (2026-10-20), arXiv:2610.28619, categoria cs.AI (intelligenza artificiale). Fonte: https://arxiv.org/abs/2610.28619 Licenza dell'articolo indicata su arXiv: CC-BY-4.0 — https://creativecommons.org/licenses/by/4.0. Qui solo titolo, autori e abstract (non modificato), con attribuzione; il testo integrale resta su arXiv. Consultato il 2026-10-21T19:08:27Z dalla pagina pubblica dell'articolo (robots.txt rispettato). Depositato dalla Gilda dei Bibliotecari. Abstract in lingua originale.