Biased Overtraining Evaluation on ArxivRollBench
0.22RSIIPhi-1
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Phi-12025.07 | 0.22 | 5.21 | |
| Llama3-8BParameters=8B2025.07 | 0.45 | 2 | |
| Phi-1.52025.07 | 0.5 | 4.02 | |
| Qwen2-7BParameters=7B2025.07 | 0.51 | 1.76 | |
| Llama2-7BParameters=7B2025.07 | 0.51 | 6.2 | |
| Phi-22025.07 | 0.53 | 2.39 | |
| Phi-3.5-miniModel size=mini2025.07 | 0.57 | 3.27 | |
| Qwen2.5-72BParameters=72B2025.07 | 0.64 | 3.84 | |
| Qwen2.5-7BParameters=7B2025.07 | 0.66 | 2.21 | |
| Phi-3-miniModel size=mini2025.07 | 0.76 | 5.84 | |
| Yi-1.5-34BParameters=34B2025.07 | 0.78 | 2.91 | |
| Llama3.1-8BParameters=8B2025.07 | 0.96 | 3.46 | |
| Llama3.1-70BParameters=70B2025.07 | 1.19 | 3.66 | |
| Llama-3.1-Nemotron-70BParameters=70B2025.07 | 1.3 | 3.88 |