AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

GTR-Turbo: Merged Checkpoint is Secretly a Free Teacher for Agentic VLM Training

Tong Wei, Yijun Yang, Changhao Zhang, Junliang Xing, Yuanchun Shi, Zongqing Lu, Deheng Ye

CVPR 2026 · Reinforcement learning › RL from feedback · approach: Scaling RL with verifiable rewards · 2 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 GPT-4o corrector with merged-checkpoint teacher (KL guidance)GTR (Qwen2.5-VL-7B) · Points24success rate (%): 44.5 -> 53.5Improved
use SFT guidance from merged teacher instead of GPT-4o correctorGTR (Qwen2.5-VL-7B) · Points24success rate (%): 44.5 -> 48.0Improved
use soft-logit KL guidance instead of SFT guidanceGTR-Turbo · Points24success rate (%): 48.0 -> 53.5Improved
use KL guidance instead of GPT-4o corrector (training time)GTR · Points24training time (hours): 191h -> 89hImproved
use KL guidance instead of GPT-4o corrector (cost)GTR · Points24additional cost (USD): $307.78 -> $114.81Improved
use merged-checkpoint KL teacher instead of GPT-4o correctorGTR · ALFWorldaverage success rate: 0.16 -> 0.15No clear effect
use SFT guidance from merged teacher instead of GPT-4o correctorGTR · ALFWorldaverage success rate: 0.16 -> 0.12Hurt
use merged teacher instead of static base-model KL reference (ablation)GTR-Turbo · Points24success rateImproved
use RL exploration instead of rejection sampling (ablation)self-improvement baseline · Points24success rateImproved
guide full response (thought + action) instead of thought only (ablation)GTR-Turbo · Points24success rateHurt
use TIES merging instead of linear averaging (ablation)GTR-Turbo · Points24success rateImproved
clip negative KL estimates instead of raw K1 (ablation)GTR-Turbo (KL) · Points24success rateImproved

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