# Spec-AUF: Accept-Until-Fail Training under Train-Inference Misalignment for Masked Block Drafters

arXiv: https://arxiv.org/abs/2607.01893 · HTML: https://mlsanity.com/paper/2607.01893/

## Key facts

- Cited by 3 papers in the map (data through 14 Oct 2026).
- Placed in Speculative decoding, one of 12 directions in Efficient LLM inference and serving.
- Builds on 14 of its 20 references that are in the map.
- First posted on arXiv 2 Jul 2026.

## Placed in

- Efficient LLM inference and serving / [Speculative decoding](https://mlsanity.com/inference/speculative-decoding/)

## Builds on (most cited first, 14 of 14)

- [Fast Inference from Transformers via Speculative Decoding](https://mlsanity.com/paper/2211.17192/) (2022)
- [Accelerating Large Language Model Decoding with Speculative Sampling](https://mlsanity.com/paper/2302.01318/) (2023)
- [EAGLE-3: Scaling up Inference Acceleration of Large Language Models via Training-Time Test](https://mlsanity.com/paper/2503.01840/) (2025)
- [DFlash: Block Diffusion for Flash Speculative Decoding](https://mlsanity.com/paper/2602.06036/) (2026)
- [How (not) to Train your Generative Model: Scheduled Sampling, Likelihood, Adversary?](https://mlsanity.com/paper/1511.05101/) (2015)
- [Domino: Decoupling Causal Modeling from Autoregressive Drafting in Speculative Decoding](https://mlsanity.com/paper/2605.29707/) (2026)
- [SpecDiff-2: Scaling Diffusion Drafter Alignment For Faster Speculative Decoding](https://mlsanity.com/paper/2511.00606/) (2025)
- [From Next-Token to Next-Block: A Principled Adaptation Path for Diffusion LLMs](https://mlsanity.com/paper/2512.06776/) (2025)
- [LK Losses: Direct Acceptance Rate Optimization for Speculative Decoding](https://mlsanity.com/paper/2602.23881/) (2026)
- [D-PACE: Dynamic Position-Aware Cross-Entropy for Parallel Speculative Drafting](https://mlsanity.com/paper/2605.18810/) (2026)
- [GRIFFIN: Effective Token Alignment for Faster Speculative Decoding](https://mlsanity.com/paper/2502.11018/) (2025)
- [DreamReasoner-8B: Block-Size Curriculum Learning for Diffusion Reasoning Models](https://mlsanity.com/paper/2606.19257/) (2026)
- [Teaching Diffusion to Speculate Left-to-Right](https://mlsanity.com/paper/2606.11552/) (2026)
- [Variational Speculative Decoding: Rethinking Draft Training from Token Likelihood to Sequence Acceptance](https://mlsanity.com/paper/2602.05774/) (2026)

## Built on by (most cited first, 3 of 3)

- [DFlow: Enabling Verifier Information Flow in Block Diffusion Speculative Decoding](https://mlsanity.com/paper/2609.06498/) (2026)
- [TreeSpark: Calibrated, Load-Adaptive Draft Trees for Semi-Autoregressive Speculative Decoding](https://mlsanity.com/paper/2609.22098/) (2026)
- [Diffusion Drafts, AR Verifies: Accelerating Document OCR with Self-Speculative Decoding](https://mlsanity.com/paper/2609.26638/) (2026)

## Abstract (arXiv)

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.