AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

TMS: Trajectory-Mixed Supervision for On-Policy Self Distillation

Rana Shahroz, Zijie Liu, Zhen Tan, Charles Fleming, Tianlong Chen

ICML 2026 · Reinforcement learning › RL from feedback · approach: RL for LLM reasoning

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
mix supervision across the training trajectory instead of rewriting labels once (add trajectory mixing vs Rephrase-SFT) (ablation)TMS · MATH-trained, Qwen-2.5-3B-Instruct, ARC-Caccuracy (%): 60.6 -> 79.2Improved
add trajectory mixing vs base-snapshot T-matched controls (ablation)TMS · MATH-trained, Qwen-2.5-3B-Instruct, ARC-Caccuracy (%): 64.7 -> 79.2Improved
add trajectory mixing vs final-snapshot T-matched controls (ablation)TMS · MATH-trained, Qwen-2.5-3B-Instruct, ARC-Caccuracy (%): 61.5 -> 79.2Improved
add trajectory mixing vs base-snapshot controls (ablation)TMS · MATH-trained, Qwen-2.5-3B-Instruct, cross-taskCross-task Avg delta (lower better): 14.7 -> 2.3Improved
add oracle label mixing (alpha 0.25 vs 0.0) (ablation)TMS · MATH-trained, Qwen-2.5-3B-Instruct, MATHaccuracy (%): 80.6 -> 81.1Improved
mix oracle labels at alpha 0.25 rather than oracle-only SFT (ablation)TMS · MATH-trained, Qwen-2.5-3B-Instruct, ARC-Caccuracy (%): 42.7 -> 79.2Improved
sample checkpoints uniformly rather than early-only (ablation)TMS · MATH-trained, Qwen-2.5-3B-Instruct, ARC-Caccuracy (%): 68.1 -> 79.2Improved
sample checkpoints uniformly rather than late-only (ablation)TMS · MATH-trained, Qwen-2.5-3B-Instruct, ARC-Caccuracy (%): 70.3 -> 79.2Improved
sample checkpoints uniformly rather than mid-only (ablation)TMS · MATH-trained, Qwen-2.5-3B-Instruct, ARC-Caccuracy (%): 78.4 -> 79.2No clear effect
use TMS instead of standard SFTQwen-2.5-1.5B-Instruct · ARC-C retentionaccuracy (%): 42.3 -> 65.4Improved
use TMS instead of standard SFTQwen-2.5-1.5B-Instruct · MATH Pass@100Pass@100 (%): 66.0 -> 74.0Improved
use TMS instead of standard SFTLLaMA-3.1-8B-Instruct · ARC-C retentionaccuracy (%): 53.2 -> 83.1Improved

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