Metraipolis › archive › dc_aaa6275ae93b
UBTree: Parallel Tree Drafting via Unigram and Bigram Models for Speculative Decoding — arXiv:2609.39972
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.39972
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-07T13:14:02.763Z
- SHA-256 of the archived text
d753998e0e8c8578fe6209926d04f5890d0a72bc0b7c0e50da64cbc2ecb37423- Stable URL of this page
https://metraipolis.com/documento/dc_aaa6275ae93b
How to cite
UBTree: Parallel Tree Drafting via Unigram and Bigram Models for Speculative Decoding — arXiv:2609.39972 — arXiv (en), https://arxiv.org/abs/2609.39972 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_aaa6275ae93b (Metraipolis archive dc_aaa6275ae93b, sha256 d753998e0e8c8578fe6209926d04f5890d0a72bc0b7c0e50da64cbc2ecb37423)
BibTeX
@misc{dcaaa6275ae93b,
title = {UBTree: Parallel Tree Drafting via Unigram and Bigram Models for Speculative Decoding — arXiv:2609.39972},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.39972},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_aaa6275ae93b; sha256 d753998e0e8c8578fe6209926d04f5890d0a72bc0b7c0e50da64cbc2ecb37423},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.39972. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Speculative decoding accelerates language model inference by verifying multiple draft tokens in a single target-model pass. Recent parallel drafters have achieved breakthrough performance in frontier production models, but their effectiveness deteriorates as the entropy of target distributions increases due to insufficient draft diversity. To overcome this bottleneck without sacrificing parallelism, we introduce UBTree, a parallel drafter that couples a Unigram proposer with a Bigram selector to construct drafting Trees. The unigram proposer is trained with the standard cross-entropy objective to generate candidate tokens independently for each position, while a lightweight bigram selector predicts transition scores between adjacent candidate pairs. Unlike the proposer, the selector is trained with a renormalized KL objective on high-temperature data. This tree-native training broadens the supervision beyond the greedy path, encouraging plausible alternative branches that improve the chance of accepting additional tokens during tree verification. Across seven standardized benchmarks with Qwen3-4B and Qwen3-8B, UBTree achieves an average speedup of $5.84$--$6.94\times$ over autoregressive decoding and outperforms DARTree in all 28 comparisons. Production-scale evaluation further demonstrates UBTree's advantage over frontier baselines such as DSpark. 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). Pagina del lavoro: https://arxiv.org/abs/2609.39972 — 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 2026-10-07T12:44:51.873Z 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).