List your company

Visual Policy Reasoning on ClevrPolicy-T (N=4)

99.15AccuracyTriMPI w/ PoRo-GRPO
22.7142.55562.482.245Oct 10, 2025

Evaluation results

MethodLinks
TriMPI w/ PoRo-GRPOComplexity=N=42025.1099.15
CoT SFT + GRPOComplexity=N=42025.1098.25
TriMPI w/ PoRo-DAPOComplexity=N=42025.1097.3
CoT SFT + DAPOComplexity=N=42025.1097.15
CoT SFTComplexity=N=42025.1049.1
In-ContextComplexity=N=42025.1032.2
Direct SFTComplexity=N=42025.1025.65