Reasoning on Winogrande 5-shot
0.749Normalized Log AccuracyHATified-SFT
Evaluation results
| Method | Links | ||
|---|---|---|---|
| HATified-SFTshots=5-shot2026.03 | 0.749 | — | |
| Llama-Instructshots=5-shot2026.03 | 0.697 | — | |
| HATifiedshot count=5-shot, DPO evaluation=true2026.03 | 0.687 | 5.16 | |
| Llama-3.1-8B-TFree-HAT-SFTshots=5, SFT=true2026.03 | 0.677 | 5.16 | |
| T-Freeshot count=5-shot, DPO evaluation=true2026.03 | 0.676 | 5.16 | |
| Llamashot count=5-shot, DPO evaluation=true, backbone=Llama-3.1-8B-Instruct2026.03 | 0.673 | 4.91 | |
| Tülushot count=5-shot, DPO evaluation=true, backbone=Llama-3.1-Tulu-3-8B2026.03 | 0.635 | 4.91 | |
| Llama-3.1-Tulu-3-8B-SFTshots=5, SFT=true2026.03 | 0.602 | 4.91 |