AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

Set Diffusion: Interpolating Token Orderings between Autoregression and Diffusion for Fast and Flexible Decoding

Marianne Arriola, Volodymyr Kuleshov

ICML 2026 · Image generation › Fast diffusion sampling · approach: Masked diffusion language model decoding · 0 citations (OpenAlex)

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 efficient token-level training (all L conditionals per ordering) (ablation)SW-SetDLM S<=L · LM1Btest perplexity: 54.49 -> 42.02Improved
add efficient token-level training (ablation)SW-SetDLM S<=L · LM1Bgradient variance: 7.50 x 10^-2 -> 2.73 x 10^-2Improved
add causal mask (KV-cache-compatible architecture) (ablation)SW-SetDLM S<=L · LM1Btest perplexity: 46.00 -> 42.02Improved
add causal mask (ablation)SW-SetDLM S<=L · LM1Bgradient variance: 15.01 x 10^-2 -> 2.73 x 10^-2Improved
add efficient training and causal mask with KV caching (ablation)SW-SetDLM S<=L · GSM8K (order-agnostic schedule)0-shot pass@1: 6.37 -> 26.61Improved
add efficient training and causal mask with KV caching (ablation)SW-SetDLM S<=L · GSM8K (order-agnostic schedule)decoding throughput: 24.48 -> 40.83Improved
replace block diffusion S=16 with set diffusion S<=32BD3LM · GSM8K0-shot pass@1: 50.49 -> 61.94Improved
replace block diffusion S=8 with set diffusion S<=16BD3LM · GSM8Kdecoding throughput: 54.67 -> 56.50Improved
replace block diffusion S=4 with set diffusion S<=8BD3LM · GSM8K0-shot pass@1: 63.53 -> 66.41Improved
replace block diffusion S=4 with set diffusion S<=8BD3LM · GSM8Kperplexity: 1.41 -> 1.42No clear effect
replace block diffusion with set diffusion S<=32BD3LM S=16 · ROCStories infilling 1/5ROUGE-1: 9.2 -> 11.6Improved
replace block diffusion with set diffusion S<=32BD3LM S=16 · ROCStories infilling 1/5decoding throughput: 105.8 -> 132.0Improved

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