AI Paper Atlas

Research map / Large language models / Reasoning and chain of thought

InfoDLM: an Information-Adaptive Framework for Discrete Diffusion Language Model Pretraining

Shirou Jing, Chunshu Wu, Chuan Liu, Arghavan Bahadorinejad, Feitong Qiao, Dongfang Liu, Tony Geng

ICML 2026 · Large language models › Reasoning and chain of thought · approach: Test-time compute scaling

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes InfoDLM, an information-adaptive masking framework for discrete diffusion language model pretraining that targets informative tokens instead of heuristic random masking.

Problem

Discrete diffusion language models (DLMs) are pretrained with heuristic random masking, which does not focus on the most informative tokens, so compute is wasted on redundant or trivial tokens.

Approach

Replaces random masking in mask-and-denoise DLM pretraining with an adaptive, information-based masking scheme that selects which tokens to mask according to their informativeness. Full method details are unavailable beyond the abstract.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
replace random masking with a policy trained on the TIG reward (ablation)SMDM 472M · GSM8K, 20B tokensaccuracy: 40.29 -> 50.8Improved
add cyclic (alternating) policy/DLM training without phase conditioning (ablation)SMDM + TIG reward · GSM8K, 35B tokensaccuracy: 51.07 -> 50.61No clear effect
add phase signal via FiLM conditioning of mask policy (ablation)SMDM + TIG reward + cyclic training · GSM8K, 20B tokensaccuracy: 48.52 -> 51.78Improved
add phase signal via FiLM conditioning of mask policy (ablation)SMDM + TIG reward + cyclic training · ANLI, 35B tokensaccuracy: 46.70 -> 49.30Improved
use learned InfoDLM masking instead of random masking (ablation)SMDM 472M · ANLI, 35B tokensaccuracy: 35.50 -> 49.30Improved
use span masking instead of random masking (ablation)SMDM 472M · GSM8K, 35B tokensaccuracy: 40.48 -> 41.42No clear effect
use PMI or frequency heuristic masking instead of random masking (ablation)SMDM 472M · GSM8K, 35B tokensaccuracy: 40.48 -> 42.54 (PMI)Improved
InfoDLM at 10B tokens versus baselines at 35B tokens (ablation)InfoDLM 472M · ANLIaccuracy: 41.20 vs 39.60 (PMI, 35B)Improved
InfoDLM instead of SMDM, zero-shot HellaSwag at 472M472M model · HellaSwag (zero-shot)accuracy: 30.04 -> 27.99Hurt
InfoDLM instead of SMDM at 1B scale, GSM8K SFT1B model · GSM8K (SFT)accuracy: 57.09 -> 57.24No clear effect
InfoDLM instead of SMDM at 1B scale, zero-shot1B model · HellaSwag (zero-shot)accuracy: 36.48 -> 42.37Improved
alternating MPN updates add compute overheadInfoDLM 472M · per 5:1 cycletime per cycle (s): 10.10 -> 11.52Hurt

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