Cloze task on SSH Cloze
81.8Raw AccuracyPhi-4
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Phi-4Size=14B, Training tokens=9.8T2026.04 | 81.8 | 81.8 | |
| OLMO-2Size=7B, Training tokens=4T2026.04 | 78.2 | 76.4 | |
| SHARESize=14B, Training tokens=96B2026.04 | 77.1 | 79.6 | |
| OLMO-2-Step-20kSize=13B, Training tokens=168B2026.04 | 74.9 | 73.8 | |
| Phi-4Size=4B, Training tokens=5T2026.04 | 73.8 | 69.8 | |
| SHARESize=4B, Training tokens=28B2026.04 | 69.8 | 66.2 | |
| SciBERTSize=110M, Training tokens=3B2026.04 | 67.9 | 62.9 | |
| PythiaSize=12B, Training tokens=300B2026.04 | 67.3 | 61.5 | |
| SSCI-SciBERT-e2Size=110M, Training tokens=1B*2026.04 | 66.9 | 67.6 | |
| PythiaSize=3B, Training tokens=300B2026.04 | 65.8 | 63.6 | |
| BERTSize=110M, Training tokens=5B*2026.04 | 58.2 | 57.5 |