List your company

Instruction hierarchy resolution on 5-level evaluation suite 1.0 (full)

83.8PPA (Macro)GW-DPO (bi.)
22.12838.13954.1570.161Jun 9, 2026

Evaluation results

MethodLinks
GW-DPO (bi.)Margin Schedule=bilateral, Levels=52026.0683.888.579.30.0440.0845.74.41100100
DPOMargin Schedule=standard2026.0681.485.2790.0240.053124.3100100
3-level (lin.)Margin Schedule=linear, Levels=32026.0677.882.6780.0020.0847.44.26100100
GW-DPO (lin.)Margin Schedule=linear, Levels=52026.0671.975.184.30.1250.1432.43.88100100
No-ISEMargin Schedule=linear, Levels=5, ISE Component=false2026.0667.772.8810.1330.14957.63.75100100
SFTConfiguration=SFT2026.0633.238.849.30.1610.2149.22.4248.352.4
baselineConfiguration=baseline2026.0624.527.1660.4150.4154.42.1549.156.3