Block drafters propose several tokens in one forward pass, before earlier target tokens are realised. Their rejection mixes two losses: missing within-block path information and imperfect modelling of observable information. Accepted length cannot distinguish them. We separate the two with an information floor, the minimum expected rejection at a specified conditioning order; rejection above this floor is the model gap. Estimating both from target rollouts across four domains, four open-weight targets, and a frontier API target yields three findings. First, the all-parallel floor reaches $0.286$ at the final slot on Qwen3-4B, limiting even the best proposal to $71\%$ per-slot acceptance. Second, one realised token removes $86$--$100\%$ of this floor, a locality also recovered by an independent mutual-information analysis. Third, current drafters remain far above their floors: the final-slot model gap accounts for $43$--$64\%$ of DFlash rejection and $85$--$92\%$ of DSpark's oracle-conditioned rejection. These findings separate the value of short-range conditioning from proposal quality.
This work measures the boundary choice of a test-time controller to choose between stopping at the cap (strict) and allowing the current attempt to finish (advisory) with paired offline replays of public traces of AIME, BrUMO and HMMT problems and two archive configurations of one model.
Gui-Lin Zhang, Zi-Qi Tan, Wu-Lan Guo et al.· 0 citations
When an encoded attack passes a content guard, the guard either never represented the payload's harmful content or represented it and failed to act. End-to-end attack success rate reports one number for both, yet the two have opposite remedies: one is a representational limit that more safety training cannot reach, the...
Haoyu Zhang, Yi Feng, Shi-Bo Zheng et al.· 0 citations
Block diffusion speculative decoding improves LLM inference efficiency by proposing a block of future tokens in parallel and verifying them with a single forward pass through the target model. However, existing methods retain only the accepted prefix and discard the rejected suffix, preventing the computation spent on...
Yao-Jie Zhang, Lin-Feng Zhang, Bin Cui et al.· 3 citations
DBloom speeds up generation with an efficient draft model (drafter) that proposes tokens for a target model to verify in one pass, preserving the target's output distribution and recommending the histogram as a preflight check before spending training compute.
This work shows that LLM-as-Judge is operationalized on cloud serving infrastructure, not of any particular model family, and proposes a minimal, low-cost reporting protocol: several judge re-runs, published stability profiles and adjacency intervals, and results under at least two judges from different families.
D-Loop is proposed, which introduces \emph{intra-block causal conditioning} within the original diffusion drafter without additional model components, and can beat DFlash and DSpark on Qwen3-4B and Qwen3-8B with obvious gains.
Ke-Cheng Chen, Yu-Yang He, Cheng Gong et al.· 0 citations
Related blog posts
MIT News · Artificial Intelligence· news.mit.eduAug 27, 2026
A new machine-learning framework aims to improve the success rate of computational protein design while moving away from results that reproduce sequences found in nature.
