Research map / Large language models / Instruction and fine-tuning
Probability-Entropy Calibration: An Elastic Indicator for Adaptive Fine-tuning
Wenhao Yu, Shaohang Wei, Jiahong Liu, Yifan Li, Minda Hu, Aiwei Liu, Hao Zhang, Irwin King
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add the entropy term (calibrate probability by expected rank) (ablation) | RankTuner (Qwen2.5-Math-7B) · AIME24 Pass@k | Pass@k | Improved |
| add the probability term (ablation) | RankTuner (Qwen2.5-Math-7B) · AIME24 Pass@k | Pass@1 / Pass@k | Mixed |
| use dynamic exponent K(xi) instead of fixed exponents (ablation) | RankTuner (Qwen2.5-Math-7B) · math reasoning | Pass@k | Improved |
| use RankTuner instead of Alpha Power loss (alpha=0.5) | Qwen2.5-Math-7B · AIME24 / OlympiadBench | Pass@16 | Improved |
| use RankTuner instead of entropy regularization (alpha=0.02) | Qwen2.5-Math-7B · AIME24 / OlympiadBench | Pass@16 | Improved |
| apply RankTuner reweighting on top of base model | Qwen2.5-Math-7B · MATH-OAI | Pass@1: 31.79 -> 68.60 | Improved |
| apply RankTuner reweighting on top of base model | Qwen3-8B · MATH-OAI | Pass@1: 65.14 -> 72.38 | Improved |
| plain SFT instead of the original model | Qwen3-8B · MATH-OAI | Pass@1: 65.14 -> 54.83 | Hurt |
| use RankTuner instead of TALR | Qwen3-8B · Minerva Math | Pass@1: 40.46 -> 38.26 | Hurt |
| use RankTuner instead of best baseline | Qwen2.5-Math-7B · AMC23 | Pass@16: 82.50 -> 82.50 | No clear effect |
| use RankTuner instead of DFT | Qwen2.5-Math-7B · ARC-C and GPQA (OOD) | Pass@1: 26.50 -> 53.58 (ARC-C) | Improved |
| use RankTuner instead of entropy- or prob-dominant weighting | noise diagnostic · injected noise | sequence noise hit@10%: 77% -> 9% | Improved |
Closest papers by meaning
- LIFT the Veil for the Truth: Principal Weights Emerge after Rank Reduction for Reasoning-Focused Supervised Fine-Tuning (ICML 2025 · similarity 0.86)
- The Geometry of Alignment Collapse: When Fine-Tuning Breaks Safety (NeurIPS 2026 · similarity 0.84)
- Memorize Theorems, Not Instances: Probing SFT Generalization through Mathematical Reasoning (NeurIPS 2026 · similarity 0.84)
- Mitigating Forgetting in LLM Fine-Tuning via Low-Perplexity Token Learning (NeurIPS 2025 · similarity 0.84)
- STARE: Surprisal-Guided Token-Level Advantage Reweighting for Policy Entropy Stability (NeurIPS 2026 · similarity 0.84)
- UFT: Unifying Supervised and Reinforcement Fine-Tuning (NeurIPS 2025 · similarity 0.83)
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.
