List your company

Mathematical Reasoning on MATH500 (Acc, Len, TE)

93.5AccuracySTACK
66.4673.4880.587.52Apr 10, 2026Apr 18, 2026Apr 27, 2026May 6, 2026May 15, 2026May 24, 2026Jun 2, 2026

Evaluation results

MethodLinks
STACKBackbone=DeepSeek-R1-Distill-Qwen-7B2026.0493.51,7335.39
TokenSqueezeBackbone=DeepSeek-R1-Distill-Qwen-7B2026.0492.81,9574.74
MuTISBackbone=DeepSeek-R1-Distill-Qwen-7B2026.0492.31,8874.89
PromptBackbone=DeepSeek-R1-Distill-Qwen-7B2026.0491.73,6542.51
OriginalBackbone=DeepSeek-R1-Distill-Qwen-7B2026.0491.24,0102.27
ConCISEBackbone=DeepSeek-R1-Distill-Qwen-7B2026.04912,0164.51
STACKBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.0485.31,7834.78
FP16Backbone=Phi-4-14B, K/V=16/16, bits/elem=16.0, UP (Uniform Precision)=true2026.0684.92,839—
KVarNBackbone=Phi-4-14B, K/V=2/2, bits/elem=2.3, UP (Uniform Precision)=true2026.0684.82,984—
MuTISBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.0484.61,7944.72
PromptBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.0483.74,3131.94
ConCISEBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.0483.32,5473.27
TokenSqueezeBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.0483.22,4473.4
FP16Backbone=Qwen3-4B, K/V=16/16, bits/elem=16.0, UP (Uniform Precision)=true2026.0682.63,857—
OriginalBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.0482.44,8761.69
KVarNBackbone=Qwen3-4B, K/V=2/2, bits/elem=2.3, UP (Uniform Precision)=true2026.0679.23,925—
QuaRotBackbone=Qwen3-4B, K/V=2/2, bits/elem=2.3, UP (Uniform Precision)=true2026.0678.93,907—
KittyBackbone=Qwen3-4B, K/V=2/2, bits/elem=2.4, UP (Uniform Precision)=false2026.0678.53,834—
KIVIBackbone=Qwen3-4B, K/V=2/2, bits/elem=2.3, UP (Uniform Precision)=true2026.0677.83,957—
TurboQuantBackbone=Qwen3-4B, K/V=3/3, bits/elem=4.6, UP (Uniform Precision)=true, Implementation=vLLM2026.06773,917—
QuaRotBackbone=Phi-4-14B, K/V=2/2, bits/elem=2.3, UP (Uniform Precision)=true2026.06773,029—
PolarQuantBackbone=Phi-4-14B, K/V=4/2, bits/elem=3.3, UP (Uniform Precision)=true2026.0675.83,023—
KittyBackbone=Phi-4-14B, K/V=2/2, bits/elem=2.4, UP (Uniform Precision)=false2026.0674.53,171—
KIVIBackbone=Phi-4-14B, K/V=2/2, bits/elem=2.3, UP (Uniform Precision)=true2026.0674.43,180—
TurboQuantBackbone=Phi-4-14B, K/V=3/3, bits/elem=4.5, UP (Uniform Precision)=true, Implementation=vLLM2026.0674.23,250—
KVQuant-1%Backbone=Phi-4-14B, K/V=2/2, bits/elem=2.4, UP (Uniform Precision)=false2026.0672.33,588—
PolarQuantBackbone=Qwen3-4B, K/V=4/2, bits/elem=3.3, UP (Uniform Precision)=true2026.0671.15,259—
KVQuant-1%Backbone=Qwen3-4B, K/V=2/2, bits/elem=2.4, UP (Uniform Precision)=false2026.0667.55,021—