List your company

Reasoning on HellaSwag 10-shot

81.7Normalized Log AccuracyTülu
65.89269.99674.178.204Mar 16, 2026

Evaluation results

MethodLinks
Tülushot count=10-shot, DPO evaluation=true, backbone=Llama-3.1-Tulu-3-8B2026.0381.74.66
HATified-SFTshots=10-shot2026.0380.9—
T-Freeshot count=10-shot, DPO evaluation=true2026.0378.35.34
HATifiedshot count=10-shot, DPO evaluation=true2026.0377.55.34
Llama-3.1-Tulu-3-8B-SFTshots=10, SFT=true2026.0377.34.66
Llamashot count=10-shot, DPO evaluation=true, backbone=Llama-3.1-8B-Instruct2026.0376.44.66
Llama-3.1-8B-TFree-HAT-SFTshots=10, SFT=true2026.0374.75.34
Llama-Instructshots=10-shot2026.0366.5—