List your company

Common-sense Reasoning on ARC-C, ARC-E, HellaSwag, Lambada, PIQA, and WinoGrande (test)

44.88ARC-C AccuracyRAT+
43.278443.694244.1144.5258Feb 20, 2026

Evaluation results

MethodLinks
RAT+Pre-training Tokens=200BT, D=2, L*=64, T=<= 3002026.0244.8876.2665.1553.4875.4160.362.58
RAT+Pre-training Tokens=200BT, D=4, L*=64, T=<= 3002026.0244.8876.1465.1753.1375.360.2262.47
RAT+Pre-training Tokens=100BT, D=4, L*=64, T=<= 3002026.0244.7175.8462.6550.7973.9959.5161.25
RAT+Pre-training Tokens=200BT, D=1, L*=64, T=<= 3002026.0244.6276.3565.2854.1875.2460.5462.7
RAT+Pre-training Tokens=100BT, D=1, L*=64, T=<= 3002026.0244.275.4262.8552.1474.3760.5461.59
RAT+Pre-training Tokens=200BT, D=16, L*=64, T=<= 3002026.0244.1175.9364.3652.0575.6858.6461.8
RAT+Pre-training Tokens=200BT, D=64, L*=64, T=<= 3002026.0244.1175.9763.4848.5275.5758.8761.09
RAT+Pre-training Tokens=100BT, D=2, L*=64, T=<= 3002026.0244.0375.3462.6751.5274.3259.4361.22
RAT+Pre-training Tokens=100BT, D=16, L*=64, T=<= 3002026.0244.0375.6362.3449.9574.158.3360.73
RAT+Pre-training Tokens=200BT, D=8, L*=64, T=<= 3002026.0243.9475.8464.7152.8675.4658.861.94
RAT+Pre-training Tokens=200BT, D=32, L*=64, T=<= 3002026.0243.8675.863.8150.975.7358.1761.38
RAT+Pre-training Tokens=100BT, D=64, L*=64, T=<= 3002026.0243.7775.5161.3545.6274.2756.5959.52
RAT+Pre-training Tokens=200BT, D=128, L*=64, T=<= 3002026.0243.6976.2662.9946.575.6858.1760.55
RAT+Pre-training Tokens=100BT, D=8, L*=64, T=<= 3002026.0243.5275.8862.6150.2874.2758.4860.84
RAT+Pre-training Tokens=100BT, D=32, L*=64, T=<= 3002026.0243.3475.2562.0148.0173.9957.3860
RAT+Pre-training Tokens=100BT, D=128, L*=64, T=<= 3002026.0243.3475.6761.0843.5374.2157.3859.2