AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Hierarchy Decoding: A Training-free Parallel Decoding Strategy for Diffusion Large Language Models

Xiaojing Qi, Lun Du, Xinyuan Zhang, Lanning Wei, Tao Jin, Da Zheng

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Parallel and early-exit decoding

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use Hierarchy-dLLM hierarchical decoding instead of Fast-dLLMLLaDA-Instruct-8B · GSM8KTPS: 12.80 -> 19.41Improved
use Hierarchy-dLLM instead of WINOLLaDA-Instruct-8B · GSM8Kscore (accuracy): 68.75 -> 77.18Improved
use Hierarchy-dLLM instead of Fast-dLLMDream-7B · Math500score (accuracy): 12.0 -> 16.0Improved
use Hierarchy-dLLM instead of vanilla decodingDream-7B · GSM8Kscore (accuracy): 75.8 -> 73.2Hurt
use Hierarchy-dLLM instead of vanilla decodingLLaDA-Instruct-8B · MBPPTPS: 0.65 -> 11.20Improved
increase generation length from 256 to 1024 (ablation)Hierarchy-dLLM · GSM8Kspeedup over vanilla TPS: 4.49x -> 11.92xImproved
increase block length from 16 to 64 (ablation)Hierarchy-dLLM · GSM8KTPS speedup: 5.81x -> 8.68xImproved
increase block length from 32 to 64 (ablation)Hierarchy-dLLM · GSM8Kscore (accuracy): 83.70 -> 81.35Hurt
decrease block length from 32 to 16 (ablation)Hierarchy-dLLM · GSM8Kscore (accuracy): 83.70 -> 81.58Hurt
raise the high threshold tau_high (ablation)Hierarchy-dLLM · GSM8KTPSHurt
vary the low threshold tau_low (ablation)Hierarchy-dLLM · GSM8KTPSNo clear effect
use sparse masking instead of consecutive masking (ablation)LLaDA (preliminary study) · HumanEval and GSM8Kapproximate KL divergenceImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy