Reasoning on ARC Easy 25-shot
92Normalized Log AccuracyHATified-SFT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| HATified-SFTshots=25-shot2026.03 | 92 | — | |
| Llama-Instructshots=25-shot2026.03 | 91.1 | — | |
| HATifiedshot count=25-shot, DPO evaluation=true2026.03 | 89.7 | 5.53 | |
| T-Freeshot count=25-shot, DPO evaluation=true2026.03 | 89.4 | 5.53 | |
| Llama-3.1-8B-TFree-HAT-SFTshots=25, SFT=true2026.03 | 88.9 | 5.53 | |
| Llamashot count=25-shot, DPO evaluation=true, backbone=Llama-3.1-8B-Instruct2026.03 | 88 | 4.94 | |
| Tülushot count=25-shot, DPO evaluation=true, backbone=Llama-3.1-Tulu-3-8B2026.03 | 81.6 | 4.94 | |
| Llama-3.1-Tulu-3-8B-SFTshots=25, SFT=true2026.03 | 79.9 | 4.94 |