List your company

Math on GSM8K (Speedup and τ)

8.08SpeedupBASTION
2.96324.29165.626.9484May 28, 2026Jun 3, 2026Jun 10, 2026Jun 17, 2026Jun 23, 2026Jun 30, 2026Jul 7, 2026

Evaluation results

MethodLinks
BASTIONModel=Qwen3-4B, Temperature=1, Hardware=H1002026.058.088.78
BASTIONModel=Qwen3-8B, Temperature=0, Hardware=H1002026.057.649.02
BASTIONModel=Qwen3-8B, Temperature=1, Hardware=H1002026.057.488.98
BASTIONModel=Qwen3-4B, Temperature=0, Hardware=H1002026.057.268.91
DFlashModel=Qwen3-8B, Temperature=0, Hardware=H100, Block size=162026.055.876.55
DFlashModel=Qwen3-4B, Temperature=0, Hardware=H100, Block size=162026.055.836.51
DFlashModel=Qwen3-4B, Temperature=1, Hardware=H100, Block size=162026.055.555.98
DFlashModel=Qwen3-8B, Temperature=1, Hardware=H100, Block size=162026.055.095.82
DFlash-TfMTemperature=1.0, Reasoning Mode=Reasoning, Path=Tree, Budget=642026.074.8510.63
DFlash-TfMTemperature=0.0, Reasoning Mode=No Reasoning, Path=Tree, Budget=642026.074.6510.56
EAGLE-3Model=Qwen3-4B, Temperature=1, Hardware=H100, Tree budget size=602026.053.392.75
EAGLE-3Model=Qwen3-4B, Temperature=0, Hardware=H100, Tree budget size=602026.053.232.8
EAGLE-3Model=Qwen3-8B, Temperature=1, Hardware=H100, Tree budget size=602026.053.222.61
EAGLE-3Model=Qwen3-8B, Temperature=0, Hardware=H100, Tree budget size=602026.053.162.69