Robustness Evaluation on HellaSwag SAGE-generated
74.17Overall Accuracy (OA)GPT-4o
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| GPT-4oFamily=GPT2026.05 | 74.17 | 65.68 | 11.44 | 50.67 | |
| Qwen3.5-35B-A3BFamily=Qwen3.52026.05 | 71.79 | 37.6 | 34.2 | 29.26 | |
| Llama-3.1-70BFamily=Llama2026.05 | 71.54 | 35.68 | 35.86 | 27.93 | |
| Qwen3.5-9BFamily=Qwen3.52026.05 | 66.21 | 36.73 | 29.48 | 26.86 | |
| Gemma-3-27BFamily=Gemma-32026.05 | 66.17 | 34.41 | 31.76 | 25.33 | |
| Llama-3.1-8BFamily=Llama2026.05 | 64.71 | 34.11 | 30.6 | 24.79 | |
| Gemma-3-12BFamily=Gemma-32026.05 | 64.5 | 33.73 | 30.77 | 24.33 | |
| Gemma-3-4BFamily=Gemma-32026.05 | 60.79 | 33.39 | 27.4 | 23.15 | |
| Qwen3.5-4BFamily=Qwen3.52026.05 | 60.5 | 34.73 | 25.77 | 23.83 | |
| Llama-3.2-3BFamily=Llama2026.05 | 59.54 | 32.84 | 26.7 | 22.57 | |
| Qwen3.5-2BFamily=Qwen3.52026.05 | 54.12 | 32.5 | 21.62 | 20.57 | |
| Llama-3.2-1BFamily=Llama2026.05 | 53.67 | 31.83 | 21.83 | 20.29 |