Reasoning on BBH (Normalized Log Accuracy and Compression)
70.6Normalized Log AccuracyLlama-Instruct
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Llama-InstructShots=3-shot2026.03 | 70.6 | — | |
| HATified-SFTShots=3-shot2026.03 | 65.2 | — | |
| Llamashots=3-shot2026.03 | 52.6 | 3.79 | |
| HATifiedshots=3-shot2026.03 | 50 | 4.67 | |
| T-Freeshots=3-shot2026.03 | 48.6 | 4.67 | |
| Tülushots=3-shot2026.03 | 46.2 | 3.79 |