Research map / Large language models / Instruction and fine-tuning
Anchored Supervised Fine-Tuning
He Zhu, Junyou Su, Peng Lai, Ren Ma, Wenjia Zhang, Linyi Yang, Guanhua Chen
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 forward-KL anchoring to DFT (ablation) | DFT · Medical avg, 10k, LLaMA-2-7B | accuracy: 29.19 -> 42.03 | Improved |
| add forward-KL anchoring to DFT (ablation) | DFT · Math avg, 100k, Qwen2.5-7B | accuracy: 26.04 -> 30.50 | Improved |
| add forward-KL anchoring to DFT (ablation) | DFT · Math avg, 30k, Qwen2.5-7B | accuracy: 27.66 -> 27.18 | Hurt |
| add probability reweighting (DFT) to SFT | SFT · Medical avg, 10k, LLaMA-2-7B | accuracy: 33.37 -> 29.19 | Hurt |
| add probability reweighting (DFT) to SFT | SFT · Math avg, 10k, Qwen2.5-7B | accuracy: 16.73 -> 27.77 | Improved |
| add plain KL regularization to SFT (ablation) | SFT · Medical avg, 10k, LLaMA-2-7B | accuracy: 33.37 -> 30.95 | Hurt |
| add plain KL regularization to SFT (ablation) | SFT · Math avg, 10k, Qwen2.5-7B | accuracy: 16.73 -> 16.79 | No clear effect |
| use forward KL instead of reverse KL as anchor (ablation) | ASFT · MedQA, MMLU, MedMCQA (LLaMA-2-7B) | accuracy | Improved |
| use SFT loss as anchor instead of KL, large coefficient (ablation) | ASFT-SFT · Math, alpha 0.01 -> 10.0 | accuracy: 0.275 -> 0.168 | Hurt |
| use SFT loss as anchor instead of KL (ablation) | ASFT-SFT · Medical, varying alpha | accuracy: fluctuating between 0.340 and 0.370 | Mixed |
| initialize DAPO from ASFT instead of SFT | DAPO · Medical avg, LLaMA-2-7B | accuracy: 40.24 -> 44.10 | Improved |
| use ASFT instead of SFT | SFT · Code avg, LLaMA-2-7B | pass@1 (%): 26.4 -> 27.0 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DAPO — DAPO: An Open-Source LLM Reinforcement Learning System at Scale (NeurIPS 2025) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- UFT: Unifying Supervised and Reinforcement Fine-Tuning (NeurIPS 2025 · similarity 0.92)
- Implicit Reward as the Bridge: A Unified View of SFT and DPO Connections (NeurIPS 2025 · similarity 0.89)
- Preference-Oriented Supervised Fine-Tuning: Favoring Target Model over Aligned Large Language Models (AAAI 2025 · similarity 0.89)
- AceReason-Nemotron 1.1: Advancing Math and Code Reasoning through SFT and RL Synergy (ICLR 2026 · similarity 0.89)
- Discriminative Finetuning of Generative Large Language Models without Reward Models and Human Preference Data (ICML 2025 · similarity 0.89)
- Linearization Explains Fine-Tuning in Large Language Models (NeurIPS 2025 · similarity 0.89)
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.
