{"ID":5675236,"CreatedAt":"2026-07-03T01:40:09.565152011Z","UpdatedAt":"2026-07-05T08:42:29.012495859Z","DeletedAt":null,"paper_url":"https://arxiv.org/abs/2607.01893","arxiv_id":"2607.01893","title":"Spec-AUF: Accept-Until-Fail Training under Train-Inference Misalignment for Masked Block Drafters","abstract":"Speculative decoding accelerates autoregressive generation by drafting a block of tokens that the target model verifies left-to-right, committing only the longest accepted prefix. Block (DLM-style) drafters predict the whole block in parallel, which is fast but trained with a full-block cross-entropy that supervises every position against the gold continuation -- even though inference discards every token after the first rejection. Recent acceptance-aware objectives patch this by reweighting the full-block loss; we instead use teacher-forced learning as a motivation for how supervision should concentrate on the accepted prefix. A mask-only block drafter has no input-side channel for gold-prefix conditioning, so AUF approximates that prefix-sensitive supervision on the loss side by keeping the cross-entropy support only through the drafter's first predicted failure. AUF is a single, detached change to the CE support -- no auxiliary objective, no verifier rollouts, and no change to the inference pipeline or the exactness contract. Within fixed drafter backbones and serving settings on Qwen3-8B, AUF raises the DFlash drafter's average emitted length $τ$, averaged over six benchmarks, from 2.40 to 2.61, with a gain on every benchmark, and transfers to Domino's two-branch head (2.56 to 2.68). Two findings sharpen the picture: the decay-only baseline reaches higher token accuracy on the shared block mask yet decodes worse, and on DFlash, once AUF truncates the support, the standard exponential position-decay weighting becomes empirically inert.","short_abstract":"Speculative decoding accelerates autoregressive generation by drafting a block of tokens that the target model verifies left-to-right, committing only the longest accepted prefix. Block (DLM-style) drafters predict the whole block in parallel, which is fast but trained with a full-block cross-entropy that supervises ev...","url_abs":"https://arxiv.org/abs/2607.01893","url_pdf":"https://arxiv.org/pdf/2607.01893v1","authors":"[\"Tianjian Yang\",\"Meng Li\"]","published":"2026-07-02T08:44:04Z","proceeding":"cs.AI","tasks":"[\"cs.AI\",\"cs.CL\"]","methods":"[]","has_code":false}