AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

Dual Language Models: Balancing sample-efficiency and overfitting resilience

David Samuel, Lucas Georges Gabriel Charpentier

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
add masked-diffusion objective (α=1/8 instead of pure AR) (ablation)autoregressive LM · 128 repetitions, avg 9 tasksnormalized AR score: 9.4 -> 19.1Improved
add masked-diffusion objective (α=3/4) (ablation)autoregressive LM · 32 repetitions, avg 9 tasksnormalized AR score: 22.0 -> 23.9Improved
add masked-diffusion objective (α=63/64) (ablation)autoregressive LM · 1 repetition, avg 9 tasksnormalized AR score: 26.1 -> 26.9Improved
add autoregressive objective (α=3/4) (ablation)masked-diffusion LM · 32 repetitions, PLL evalnormalized PLL score: 21.4 -> 25.3Improved
add autoregressive objective (α=1/8) (ablation)masked-diffusion LM · 128 repetitions, PLL evalnormalized PLL score: 21.0 -> 23.0Improved
evaluate dual model with prefix attention mask instead of causaldual-objective LM · 1 repetitionnormalized score: 26.9 -> 27.9Improved
evaluate dual model with prefix attention mask instead of causaldual-objective LM · 32 repetitionsnormalized score: 23.9 -> 25.0Improved
evaluate dual model with prefix attention mask instead of causaldual-objective LM · 128 repetitionsnormalized score: 19.1 -> 20.5Improved
set α so autoregressive objective sees about 16 repetitions in data-constrained trainingdual-objective LM · more than 32 repetitionsdownstream performanceImproved
use pure autoregressive training in regular-data regime (16 repetitions or fewer) (ablation)autoregressive LM · regular dataAR performanceNo clear effect
use 6 mask tokens instead of 1 in PLL (ablation)PLL evaluation · BLiMP and HellaSwagnormalized PLL score: 65.2 -> 60.3Hurt
use 6 mask tokens instead of 1 in PLL (ablation)PLL evaluation · ARC-Easy and PIQAnormalized PLL score: 18.7 -> 26.3Improved

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