List your company

Mathematical Reasoning on ASDiv (Acc, Tokens, Latency, ActRatio)

96.58AccuracyUCoT
91.369692.722394.07595.4277Oct 9, 2025

Evaluation results

MethodLinks
UCoTBackbone=Deepseek-R1-Distill-Qwen-7B, Threshold=0.52025.1096.58257.933.0155
UCoTBackbone=Deepseek-R1-Distill-Qwen-7B, Threshold=0.72025.1096.01300.123.5264
OriginalBackbone=Deepseek-R1-Distill-Qwen-7B2025.1095.47468.945.48—
UCoTBackbone=Deepseek-R1-Distill-Qwen-7B, Threshold=0.92025.1095.43384.614.5282
TokenskipBackbone=Deepseek-R1-Distill-Qwen-7B, Threshold=0.92025.1095.18431.555.0792
TokenskipBackbone=Deepseek-R1-Distill-Qwen-7B, Threshold=0.72025.1094.76337.633.9572
TokenskipBackbone=Deepseek-R1-Distill-Qwen-7B, Threshold=0.52025.1094.33301.323.5264
OriginalBackbone=Qwen3-8B2025.1093.74475.494.72—
UCoTBackbone=Qwen3-8B, Threshold=0.92025.1093.53413.674.1287
TokenskipBackbone=Qwen3-8B, Threshold=0.72025.1093.48361.373.5976
UCoTBackbone=Qwen3-8B, Threshold=0.72025.1093.32304.313.0364
UCoTBackbone=Qwen3-8B, Threshold=0.52025.1092.77257.762.5754
TokenskipBackbone=Qwen3-8B, Threshold=0.92025.1092.55446.964.4494
TokenskipBackbone=Qwen3-8B, Threshold=0.52025.1091.57294.832.9362