AI Paper Atlas

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

Diffusion LLMs Can Do Faster-Than-AR Inference via Discrete Diffusion Forcing

Xu Wang, Chenkai Xu, Yijie Jin, Jiachun Jin, Hao Zhang, Kai Yu, Zhijie Deng

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Parallel and early-exit decoding · 101 citations (OpenAlex)

Read the paper

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
apply D2F to LLaDA-InstructLLaDA-Instruct-8B · MBPP 3-shottokens per second: 0.9 -> 47.6Improved
apply D2F to LLaDA-InstructLLaDA-Instruct-8B · GSM8K 4-shotscore: 77.4 -> 77.3No clear effect
apply D2F to LLaDA-InstructLLaDA-Instruct-8B · MATH 4-shottokens per second: 21.1 -> 90.2Improved
apply D2F to Dream-BaseDream-Base-7B · GSM8K-CoT 8-shottokens per second: 9.5 -> 91.2Improved
apply D2F to Dream-BaseDream-Base-7B · GSM8K-CoT 8-shotscore: 75.0 -> 77.6Improved
add pipelined parallel decoding on top of KV cache (ablation)D2F-LLaDA · GSM8K 4-shottokens per second: 17.5 -> 52.5Improved
add pipelined parallel decoding on top of KV cache (ablation)D2F-Dream · MBPP 3-shottokens per second: 40.5 -> 105Improved
add pipelined parallel decoding on top of KV cache (ablation)D2F-Dream · MATH 4-shotscore: 35.9 -> 35.4No clear effect
use D2F progressive block noising instead of random per-block noise (ablation)D2F-Dream · MBPP 3-shotscore: 49.6 -> 54.6Improved
use D2F training instead of random noise per block (ablation)D2F-Dream · GSM8K 4-shot (tau_act 0.95)score: 77.1 -> 77.2No clear effect
use dual-state pipeline (tau_add 0.7 < tau_act) instead of tau_add = tau_act (ablation)D2F-Dream · GSM8K 4-shot (tau_act 0.85)score: 72.6 -> 74.2Improved
lower tau_add from 0.7 to 0.1 (ablation)D2F-Dream · GSM8K 4-shot (tau_act 0.85)score: 74.2 -> 75.0Improved

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