AI Paper Atlas

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

Fast-dLLM v2: Efficient Block-Diffusion LLM

Chengyue Wu, Hao Zhang, Shuchen Xue, Shizhe Diao, Yonggan Fu, Zhijian Liu, Pavlo Molchanov, Ping Luo, Song Han, Enze Xie

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
add padding to align packed samples to block size (ablation)Fast-dLLM v2 1.5B · avg of 7 suitesaverage score: 41.3 -> 42.2Improved
add complementary masking (ablation)Fast-dLLM v2 1.5B · avg of 7 suitesaverage score: 42.2 -> 45.0Improved
use block diffusion adaptation vs NTP fine-tuning on same dataQwen2.5-1.5B-Instruct · avg of 7 suitesaverage score: 44.3 -> 45.0Improved
use block diffusion adaptation vs NTP fine-tuning on same dataQwen2.5-7B-Instruct · avg of 7 suitesaverage score: 59.6 -> 60.3Improved
block diffusion vs NTP fine-tuning on MATHQwen2.5-7B-Instruct · MATHaccuracy: 72.0 -> 61.6Hurt
enable confidence-threshold 0.9 parallel decoding (ablation)Fast-dLLM v2 7B · GSM8Kthroughput (tokens/s): 39.1 -> 101.7Improved
enable confidence-threshold 0.9 parallel decoding (ablation)Fast-dLLM v2 7B · GSM8KaccuracyNo clear effect
use diffusion decoding vs AR decoding on H100 at batch 64Fast-dLLM v2 7B vs Qwen2.5-7B-Instruct · GSM8Kthroughput speedup: 1.8xImproved
use sub-block size 8 (vs 2 or 32) (ablation)Fast-dLLM v2 1.5B · HumanEvalpass@1: 38.4 -> 43.9Improved
use smaller sub-block size 2 on GSM8K (ablation)Fast-dLLM v2 1.5B · GSM8Kaccuracy: 60.2 -> 62.8Improved
change inference block size away from trained size 32 (ablation)Fast-dLLM v2 1.5B · GSM8Kaccuracy: 60.2 -> 53.2Hurt
add sub-block cache (ablation)Fast-dLLM v2 · GSM8KaccuracyNo clear effect

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