Theory Restoration on CLEGRV (test)
100Adjacent AccuracyGPT-4o
Evaluation Results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| GPT-4oDistillation Sample Size (N)=10, Number of runs=5, Timeout=600 seconds2026.06 | 100 | 99.19 | 91.73 | 94.56 | 91.91 | 100 | 98.76 | 98.6 | 100 | 100 | 95.38 | |
| MistralDistillation Sample Size (N)=10, Number of runs=5, Timeout=600 seconds2026.06 | 88.99 | 87.44 | 85.1 | — | — | — | 87.14 | — | — | — | — | |
| Gemini-3Distillation Sample Size (N)=10, Number of runs=5, Timeout=600 seconds2026.06 | 86.11 | 99.78 | 100 | 95.56 | 100 | 100 | 100 | 98.22 | 100 | 100 | 100 | |
| LLaMA3Distillation Sample Size (N)=10, Number of runs=5, Timeout=600 seconds2026.06 | 81.29 | — | — | — | — | 88.59 | — | — | — | — | — | |
| DeepSeekDistillation Sample Size (N)=10, Number of runs=5, Timeout=600 seconds2026.06 | 78.9 | 88.75 | 85.69 | 80.74 | 86.5 | 74.74 | 64.35 | 96.7 | 87.62 | 92.11 | 93.8 |