List your company

Mathematical Reasoning on GSM8K (Pass@k Metrics)

90.13Pass@1HRPO
59.34667.33875.3383.322Jun 4, 2026

Evaluation results

MethodLinks
HRPOBackbone=Qwen2.5-7B-Instruct2026.0690.1397.42
TARPOBackbone=Qwen2.5-7B-Instruct2026.0689.9497.27
GRPOBackbone=Qwen2.5-7B-Instruct2026.0689.8797.42
Pure LatentBackbone=Qwen2.5-7B-Instruct2026.0688.490.9
Entropy-RoutedBackbone=Qwen2.5-7B-Instruct2026.0688.0797.04
CoTBackbone=Qwen2.5-7B-Instruct2026.0687.9797.73
TARPOBackbone=Qwen2.5-3B-Instruct2026.0683.2396.97
HRPOBackbone=Qwen2.5-3B-Instruct2026.0682.7597.42
GRPOBackbone=Qwen2.5-3B-Instruct2026.0682.5597.57
Entropy-RoutedBackbone=Qwen2.5-3B-Instruct2026.0673.9195.53
Pure LatentBackbone=Qwen2.5-3B-Instruct2026.0673.8478.92
CoTBackbone=Qwen2.5-3B-Instruct2026.0673.596.97
TARPOBackbone=Qwen2.5-1.5B-Instruct2026.0670.7696.97
HRPOBackbone=Qwen2.5-1.5B-Instruct2026.0669.7196.66
GRPOBackbone=Qwen2.5-1.5B-Instruct2026.0669.0496.66
CoTBackbone=Qwen2.5-1.5B-Instruct2026.0660.8995.53
Entropy-RoutedBackbone=Qwen2.5-1.5B-Instruct2026.0660.6693.1
Pure LatentBackbone=Qwen2.5-1.5B-Instruct2026.0660.5365.5