AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

dTRPO : Trajectory Reduction in Policy Optimization of Diffusion Large Language Models

Wenxuan Zhang, Lemeng Wu, Changsheng Zhao, Ernie Chang, Mingchen Zhuge, Zechun Liu, DiJia Su, Hanxian Huang, Jun Chen, Chong Zhou, Raghuraman Krishnamoorthi, Vikas Chandra and 2 more

ICML 2026 · Reinforcement learning › RL from feedback · approach: RL for LLM reasoning · 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
use dTRPO trajectory-reduction DPO instead of the base modelFast-dLLM-v2-7B · GPQAaccuracy: 20.71 -> 30.30Improved
use dTRPO instead of the base modelFast-dLLM-v2-7B · MATHaccuracy: 60.26 -> 64.3Improved
use dTRPO instead of the base modelFast-dLLM-v2-7B · Arena-Hard v2.0score: 1.26 -> 7.41Improved
use dTRPO instead of VRPO (ELBO-based DPO)Fast-dLLM-v2-7B · LiveCodeBench v6accuracy: 12.89 -> 15.17Improved
use dTRPO instead of DPO with mean-field estimateFast-dLLM-v2-7B · IFEvalaccuracy: 65.62 -> 65.06No clear effect
use ratios of newly unmasked tokens (dTRPO) instead of directly maximizing masked-token probability (ablation)Fast-dLLM-v2-7B · GSM8Kaccuracy: 66.26 -> 85.97Improved
use dTRPO instead of ELBO objective on masked tokens (ablation)Fast-dLLM-v2-7B · IFEvalaccuracy: 51.02 -> 65.06Improved
mix math and code preference data with SmolTalk2 (ablation)dTRPO · LiveCodeBench v6accuracy: 13.36 -> 15.17Improved
mix math and code preference data with SmolTalk2 (ablation)dTRPO · IFEvalaccuracy: 67.10 -> 65.06Hurt
use on-policy data instead of mixed off-policy data (ablation)dTRPO · MATH and codeaccuracy: 64.30 -> 62.74 (MATH)Hurt
apply dTRPO to long-block LLaDALLaDA · MATH (256 tokens)accuracy: 27.84 -> 34.88Improved
apply dTRPO to Fast-dLLM-v2 1.5BFast-dLLM-v2 1.5B · MATHaccuracy: 37.9 -> 38.96No clear effect

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