List your company

Logic Inversion on OpenAgent Tier-3

22.7Accuracy DeltaSFT-200 + PAFT
-42.404-25.502-8.68.302Jul 1, 2026

Evaluation results

MethodLinks
SFT-200 + PAFTtraining_steps=200, PAFT=true2026.0722.7
SFT-400 + PAFTtraining_steps=400, PAFT=true2026.07-2.8
Base2026.07-8.5
SFT-800 + PAFTtraining_steps=800, PAFT=true2026.07-9.8
PAFT (alpha=0.3 default)training_steps=800, perturbation_ratio=0.32026.07-9.8
PAFT (SRP only)training_steps=800, perturbation_ratio=0.3, component=SRP2026.07-10.6
SFT-600 + PAFTtraining_steps=600, PAFT=true2026.07-10.7
PAFT (alpha=0.4)training_steps=800, perturbation_ratio=0.42026.07-11.1
PAFT (alpha=0.2)training_steps=800, perturbation_ratio=0.22026.07-15.8
PAFT (EFP only)training_steps=800, perturbation_ratio=0.3, component=EFP2026.07-27
SFT-800training_steps=8002026.07-28
PAFT (SBP only)training_steps=800, perturbation_ratio=0.3, component=SBP2026.07-31.7
SFT-400training_steps=4002026.07-32.5
SFT-600training_steps=6002026.07-33
SFT-200training_steps=2002026.07-39.9