AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

Attention Is All You Need for KV Cache in Diffusion LLMs

Quan Nguyen-Tri, Mukul Ranjan, Zhiqiang Shen

ICLR 2026 · Image generation › Fast diffusion sampling · approach: Masked diffusion language model 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
Elastic-Cache adaptive KV refresh instead of Fast-dLLM block cachingLLaDA-Instruct · GSM8K 5-shot 512accuracy (%) / throughput: 74.83 -> 77.71 acc; 44.0 -> 90.1 t/sImproved
Elastic-Cache instead of fixed-interval DeepCacheLLaDA-1.5 · GSM8K 5-shot 512accuracy / throughput: 81.4 -> 83.7 acc; 60.9 -> 139.4 t/sImproved
Raise attention threshold gamma from 0.5 to 0.95 (ablation)Elastic-Cache · GSM8K 512, LLaDA-1.5accuracy (%) vs speedup: 76.04 -> 83.02 acc; 54.9x -> 37.8x speedupMixed
Raise gamma from 0.9 to 0.95 (ablation)Elastic-Cache · GSM8K 512, LLaDA-Instructaccuracy (%): 77.71 -> 76.72Hurt
Increase generation length from 256 to 1024 (ablation)Elastic-Cache · GSM8K 5-shot, LLaDAspeedup over Fast-dLLM: 1.1x -> 6.8xImproved
Increase few-shot prefill from 3 to 8 (ablation)Elastic-Cache · GSM8K 1024, LLaDAthroughput (t/s): 185.3 -> 143.9Hurt
Use sliding window instead of block-wise decoding (ablation)Elastic-Cache · GSM8K, LLaDAaccuracyImproved
Increase sliding window size beta beyond about 64 (ablation)Elastic-Cache · GSM8K, LLaDAaccuracyHurt
More aggressive parallel decoding (4 tok/step) (ablation)Fast-dLLM · GSM8K, LLaDA-1.5accuracy (%): 80.5 -> 64.7 (1 to 4 tok/step)Hurt
More aggressive parallel decoding (4 tok/step) (ablation)Elastic-Cache · GSM8K, LLaDA-1.5accuracy (%): 82.6 -> 69.9 (1 to 4 tok/step)Hurt
Elastic-Cache on code tasks with LLaDA-1.5LLaDA-1.5 · HumanEval 0-shot 256accuracy (%): 43.29 -> 36.59Hurt
Elastic-Cache with more accurate base modelElastic-Cache · LLaDA-1.5 vs LLaDA GSM8Kspeedup: 25.2x -> 45.1x (512 tokens)Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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