AI Paper Atlas

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

SPEED: Sharpened-Teacher Distillation for Parallel Decoding of Diffusion Language Models

Qiuhong Shen, Xingyi Yang, Xinyin Ma, Gongfan Fang, Xinchao Wang

ICML 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
add sharpened-teacher distillation (SPEED) to LLaDALLaDA-8B-Instruct · GSM8K (5-shot)decoding steps / speedup: 256 steps -> 26 steps, 9.8× speedup, accuracy 79.3 -> 77.9Improved
add JSD-based Slow-Fast decoding on top of SPEEDSPEED LLaDA-8B-Instruct · MATH (4-shot)accuracy: 32.0 -> 32.8 (36 -> 39 steps)Improved
add JSD decoding on top of SPEEDSPEED LLaDA-8B-Instruct · HumanEval (0-shot)accuracy: 40.9 -> 42.1Improved
add JSD decoding on top of SPEEDSPEED Dream-7B-Instruct · GSM8K (5-shot)accuracy: 78.4 -> 78.4No clear effect
add JSD decoding on top of SPEEDSPEED Dream-7B-Instruct · MBPP (3-shot)accuracy: 55.6 -> 55.6No clear effect
add student temperature Ts>1 (ablation)SPEED distillation · GSM8K (5-shot, LLaDA)accuracy: 77.6 -> 77.9 (speedup 9.1× -> 9.8×)Improved
add teacher temperature Tt<1 (ablation)SPEED distillation · GSM8K (5-shot, LLaDA)speedup: 6.9× -> 9.8× speedup, accuracy 78.1 -> 77.9Mixed
restrict KL loss to teacher-correct positions instead of all masked tokens (ablation)SPEED distillation · GSM8K (5-shot, LLaDA)accuracy: 69.2 -> 77.9Improved
use selective KL distillation instead of entropy minimization (ablation)SPEED distillation · GSM8K (5-shot, LLaDA)accuracy: 76.1 -> 77.9 (31 -> 26 steps)Improved
use training noise level t=0.5 instead of 0.25 (ablation)SPEED distillation · GSM8K (5-shot, LLaDA)accuracy: 73.5 -> 77.9Improved
use training noise level t=0.5 instead of 1.0 (ablation)SPEED distillation · GSM8K (5-shot, LLaDA)accuracy: 72.4 -> 77.9Improved
apply SPEED trained on math data to out-of-domain tasksSPEED-LLaDA · CNN/DailyMail, TruthfulQA, QASPERquality / speedup: CNN/DM 19.7 -> 20.8 (3.4×); TruthfulQA 41.8 -> 40.6 (7.1×); QASPER 11.2 -> 10.5 (4.8×)Mixed

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