Long-form factuality evaluation on Biography
65.9AccuracyCURE
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| CUREPipeline Stage=Factuality Optimization (Full Pipeline)2026.04 | 65.9 | 0.189 | 0.268 | 0.676 | |
| SFT + F-RL + COPipeline Stage=Calibration Optimization2026.04 | 64.7 | 0.195 | 0.266 | 0.688 | |
| SFTPipeline Stage=Supervised Fine-Tuning2026.04 | 51.5 | 0.361 | 0.395 | 0.62 | |
| SFT + F-RLPipeline Stage=Feasibility RL2026.04 | 49.2 | 0.438 | 0.446 | 0.618 | |
| L2RF2026.04 | 47.1 | — | — | — | |
| Base LLMBackbone=Llama3.1-8B-Instruct2026.04 | 45.7 | 0.124 | 0.239 | 0.691 | |
| LitCab2026.04 | 45.7 | 0.125 | 0.239 | 0.693 |