AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

wd1: Weighted Policy Optimization for Reasoning in Diffusion Language Models

Xiaohang Tang, Rares Dolga, Sangwoong Yoon, Ilija Bogunovic

ICLR 2026 · Reinforcement learning › RL from feedback · approach: RL for LLM reasoning · 61 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
replace SFT + diffu-GRPO (d1) with wd1 (no SFT)LLaDA-8B-Instruct · Sudoku (gen len 256)accuracy: 17.6 -> 76.4Improved
replace d1 with wd1LLaDA-8B-Instruct · Countdown (gen len 256)accuracy: 25.8 -> 51.2Improved
replace d1 with wd1LLaDA-8B-Instruct · GSM8K (gen len 256)accuracy: 78.2 -> 80.8Improved
replace d1 with wd1LLaDA-8B-Instruct · MATH500 (gen len 512)accuracy: 38.0 -> 39.0Improved
replace d1 with wd1 (no change at length 256)LLaDA-8B-Instruct · MATH500 (gen len 256)accuracy: 34.4 -> 34.4No clear effect
remove SFT stage and use wd1 insteadd1 · training costSFT time: 2.01 hrs -> 0 hrsImproved
remove old-policy likelihood approximation (wd1 vs d1)d1 · RL training time per stepseconds/step: 103.5 -> 81.16Improved
add negative sample weight w- to wd1 (ablation)wd1 · Sudoku (gen len 256, ablation checkpoints)accuracy: 6.69 -> 25.2Improved
add negative sample weight w- to wd1 (ablation)wd1 · Countdown (gen len 256, ablation checkpoints)accuracy: 13.67 -> 51.2Improved
add negative sample weight w- to wd1 (ablation)wd1 · GSM8K (gen len 256, ablation checkpoints)accuracy: 65.66 -> 80.8Improved
add SFT before wd1 training (ablation)wd1 · Sudoku (gen len 256, ablation checkpoints)accuracy: 25.2 -> 26.5No clear effect
add SFT before wd1 training (ablation)wd1 · Countdown (gen len 256, ablation checkpoints)accuracy: 51.2 -> 43.4Hurt

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