AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

LightningRL: Breaking the Accuracy–Parallelism Trade-off of Block-wise dLLMs via Reinforcement Learning

Yanzhe Hu, Yijie Jin, Pengfei Liu, Kai Yu, Zhijie Deng

ICML 2026 · Architectures and efficiency › RNNs and normalization · approach: Distributed LLM inference systems · 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 token-level NLL loss on correct trajectories (ablation)LightningRL · GSM8Kaccuracy (%): 80.7 -> 90.3Improved
add per-reward decoupled normalization (ablation)LightningRL · GSM8Kaccuracy (%): 85.3 -> 90.3Improved
add TPF-aware filtering (dynamic sampling) (ablation)LightningRL · GSM8Kaccuracy (%): 87.2 -> 90.3Improved
add TPF-aware filtering (dynamic sampling) (ablation)LightningRL · GSM8KAUP: 454.5 -> 507.5Improved
use sequence-level policy/KL and token-level NLL reduction (Seq-Tok-Tok) instead of pure token-level (ablation)LightningRL · GSM8Kaccuracy (%): 80.0 -> 90.3Improved
use Seq-Seq-Seq reduction instead of Seq-Tok-Tok (ablation)LightningRL · GSM8KAUP: 507.5 -> 467.1Hurt
apply LightningRL RL post-training to SDAR-8B-b32SDAR-8B-b32 · average of four benchmarksAUP: 211.2 -> 437.8Improved
apply LightningRL to SDAR-8B-b32SDAR-8B-b32 · MBPPTPF: 2.44 -> 11.10Improved
apply LightningRL to SDAR-8B-b32SDAR-8B-b32 · HumanEvalaccuracy (%): 73.5 -> 72.6Hurt
apply LightningRL to SDAR-8B-b32SDAR-8B-b32 · MATH500accuracy (%): 63.6 -> 63.0No clear effect
use LightningRL instead of TraceRLSDAR-8B-b32 · MBPPTPF: 2.50 -> 11.10Improved
add decoupled reward normalization (ablation)LightningRL · GSM8K trainingcollapse ratioImproved

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