AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

BubbleSpec: Turning Long-Tail Bubbles into Speculative Rollout Drafts for Synchronous Reinforcement Learning

Yuhang Xu, Kaibin Tian, Yang Tian, Zhice Yang, Yifeng Yu, Yan Li, Shengzhong Liu, Fan Wu, Guihai Chen

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
add BubbleSpec speculative rolloutVerl rollout, Qwen3-1.7B · RL rolloutrollout time (s): 306.2 -> 167.6 (-45.2%)Improved
add BubbleSpec speculative rolloutVerl rollout, Qwen2.5-VL-7B · RL rolloutrollout time (s): 571.9 -> 400.1 (-30.0%)Improved
add BubbleSpec speculative rolloutVerl rollout, Qwen3-1.7B · RL rolloutaverage decoding steps: 6741 -> 2913 (-56.8%)Improved
add unified attention for speculative verification (ablation)BubbleSpec, Qwen3-1.7B · RL rolloutrollout time (s): 294.9s -> 167.6sImproved
use suffix tree instead of n-gram matching (ablation)BubbleSpec, Qwen3-1.7B · RL rolloutrollout time (s): 305.9 -> 167.6Improved
use suffix tree instead of n-gram matching (ablation)BubbleSpec, Qwen3-1.7B · RL rolloutacceptance length: 1.83 -> 2.15Improved
increase pre-generated samples per prompt from 16 to 32 (ablation)BubbleSpec, Qwen3-1.7B · RL rolloutrollout time (s): 167.6 -> 184.4Hurt
decrease pre-generated samples per prompt from 16 to 4 (ablation)BubbleSpec, Qwen3-1.7B · RL rolloutrollout time (s): 167.6 -> 183.8Hurt
replace Verl with BubbleSpec in RL trainingQwen2.5-VL-7B · AIME24accuracy: 80.72 -> 79.58No clear effect
replace Verl with BubbleSpec in RL trainingQwen2.5-VL-7B · Geo3K (OOD vision)accuracy: 48.25 -> 51.41Mixed
replace RollPacker with BubbleSpecQwen2.5-VL-7B · RL rolloutaverage rollout time (s): 655 -> 397Improved
exploit intra-GPU bubbles of fastest ranks in addition to inter-GPU bubbles (ablation)BubbleSpec · RL rolloutrollout stabilityHurt

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