Research map / Large language models / Retrieval and knowledge
REFO: Reinforced Evolutionary Faithfulness Optimization for Large Language Models
Yi Wang, Xiaqiang Tang, Keyu Hu, Haojie Lu, Sihong Xie
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 |
|---|---|---|---|
| apply REFO iterative self-evolution | Llama-3-8B · NQ-Swap | span EM: 73.54% -> 90.08% | Improved |
| apply REFO | Llama-3-8B · ConFiQA-MC | span EM: 33.68% -> 74.03% | Improved |
| apply REFO | Mistral-7B · Memo-Trap | span EM: 34.34% -> 54.75% | Improved |
| apply REFO | Qwen2.5-7B · FollowBench | CSL: 2.90 -> 2.92 | No clear effect |
| apply REFO | Llama-3-8B · FollowBench | CSL: 2.75 -> 2.81 | Improved |
| add attention-guided DPO loss (Att-REFO vs REFO) (ablation) | REFO Llama-3-8B · ELI5 | ROUGE-1 F1: 30.97% -> 34.03% | Improved |
| add attention-guided DPO loss (ablation) | REFO Qwen2.5-7B · ConFiQA-MC | span EM: 43.18% -> 50.38% | Improved |
| add attention-guided DPO loss (ablation) | REFO Mistral-7B · Memo-Trap | span EM: 54.75% -> 42.73% | Hurt |
| add attention-guided DPO loss (ablation) | REFO Llama-3-8B · NQ-Swap | span EM: 90.08% -> 88.22% | Mixed |
| add attention-guided DPO loss (training speed) (ablation) | REFO · avg NQ, NQ-Swap, MemoTrap | span EM per iteration | Improved |
| add relative preference optimization (Step 2) (ablation) | REFO Llama and Mistral · avg NQ, NQ-Swap, MemoTrap | span EM | Improved |
| add faithfulness score model filtering (ablation) | REFO Llama and Qwen · avg NQ, NQ-Swap, MemoTrap | span EM | Mixed |
Closest papers by meaning
- FaithEval: Can Your Language Model Stay Faithful to Context, Even If "The Moon is Made of Marshmallows" (ICLR 2025 · similarity 0.90)
- To Trust or Not to Trust? Enhancing Large Language Models' Situated Faithfulness to External Contexts (ICLR 2025 · similarity 0.89)
- SCOPE: A Self-supervised Framework for Improving Faithfulness in Conditional Text Generation (ICLR 2025 · similarity 0.88)
- ParamMute: Suppressing Knowledge-Critical FFNs for Faithful Retrieval-Augmented Generation (NeurIPS 2025 · similarity 0.88)
- HonestLLM: Toward an Honest and Helpful Large Language Model (NeurIPS 2024 · similarity 0.88)
- Walk the Talk? Measuring the Faithfulness of Large Language Model Explanations (ICLR 2025 · similarity 0.87)
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.
