List your company

Visual Policy Reasoning on ClevrPolicy-T (N=6)

77.8AccuracyTriMPI w/ PoRo-DAPO
1.8821.5941.361.01Oct 10, 2025

Evaluation results

MethodLinks
TriMPI w/ PoRo-DAPOModel Size=7B2025.1077.8
TriMPI w/ PoRo-DAPOModel Size=3B2025.1068.75
CoT SFT + DAPOModel Size=7B2025.1067.6
TriMPI w/ PoRo-GRPOModel Size=7B2025.1065.85
CoT SFT + DAPOModel Size=3B2025.1048.1
CoT SFT + GRPOModel Size=7B2025.1047.05
TriMPI w/ PoRo-GRPOModel Size=3B2025.1046.65
CoT SFT + GRPOModel Size=3B2025.1031.55
CoT SFTModel Size=3B2025.1019
CoT SFTModel Size=7B2025.1017.8
Direct SFTModel Size=7B2025.1015.15
Direct SFTModel Size=3B2025.1013.4
In-ContextModel Size=7B2025.1013.15
In-ContextModel Size=3B2025.104.8