List your company

Top-10 Attended-Token Overlap on Attention Diagnostics

94.2Top-10 Attended OverlapBlock-GTQ
31.48847.76964.0580.331Jun 23, 2026

Evaluation results

MethodLinks
Block-GTQModel=DS-R1-Qwen-7B, Bit Budget=4 b/dim2026.0694.2
Block-GTQModel=Mistral-Nemo-12B, Bit Budget=4 b/dim2026.0694.2
Block-GTQModel=DS-R1-Llama-8B, Bit Budget=4 b/dim2026.0693
Block-GTQModel=Llama-3.1-8B, Bit Budget=4 b/dim2026.0692.9
Block-GTQModel=Qwen2.5-14B, Bit Budget=4 b/dim2026.0692.6
Block-GTQModel=Qwen3-8B, Bit Budget=4 b/dim2026.0692.6
Block-GTQModel=Qwen2.5-3B, Bit Budget=4 b/dim2026.0692.5
Block-GTQModel=Qwen3-30B, Bit Budget=4 b/dim2026.0692
TQ-MSEModel=Mistral-Nemo-12B, Bit Budget=4 b/dim2026.0691.2
Block-GTQModel=DS-R1-Qwen-7B, Bit Budget=3 b/dim2026.0689.7
TQ-MSEModel=DS-R1-Llama-8B, Bit Budget=4 b/dim2026.0689.6
TQ-MSEModel=Llama-3.1-8B, Bit Budget=4 b/dim2026.0689.5
Block-GTQModel=Mistral-Nemo-12B, Bit Budget=3 b/dim2026.0689.1
TQ-MSEModel=Qwen2.5-14B, Bit Budget=4 b/dim2026.0688.3
KIVIModel=Llama-3.1-8B, Bit Budget=4 b/dim2026.0687.3
Block-GTQModel=Llama-3.1-8B, Bit Budget=3 b/dim2026.0687.2
Block-GTQModel=DS-R1-Llama-8B, Bit Budget=3 b/dim2026.0687.2
TQ-MSEModel=DS-R1-Qwen-7B, Bit Budget=4 b/dim2026.0687.2
KIVIModel=DS-R1-Llama-8B, Bit Budget=4 b/dim2026.0687
TQ-MSEModel=Qwen3-8B, Bit Budget=4 b/dim2026.0686.8
Block-GTQModel=Qwen3-8B, Bit Budget=3 b/dim2026.0686.7
TQ-MSEModel=Qwen2.5-3B, Bit Budget=4 b/dim2026.0686.4
Block-GTQModel=Qwen2.5-14B, Bit Budget=3 b/dim2026.0686.4
Block-GTQModel=Qwen2.5-3B, Bit Budget=3 b/dim2026.0686.1
Block-GTQModel=GLM-4-9B, Bit Budget=4 b/dim2026.0686
KIVIModel=DS-V2-Lite, Bit Budget=4 b/dim2026.0685.9
KIVIModel=Qwen2.5-3B, Bit Budget=4 b/dim2026.0685.7
KIVIModel=Qwen2.5-14B, Bit Budget=4 b/dim2026.0685.6
Block-GTQModel=Qwen3-30B, Bit Budget=3 b/dim2026.0685.5
KIVIModel=Qwen3-30B, Bit Budget=4 b/dim2026.0685.5
KIVIModel=DS-R1-Qwen-7B, Bit Budget=4 b/dim2026.0685.1
KIVIModel=Llama-3.1-8B, Bit Budget=3 b/dim2026.0685
KIVIModel=DS-R1-Llama-8B, Bit Budget=3 b/dim2026.0684.6
TQ-MSEModel=Mistral-Nemo-12B, Bit Budget=3 b/dim2026.0684.2
TQ-MSEModel=Qwen3-30B, Bit Budget=4 b/dim2026.0684.1
KIVIModel=Qwen3-30B, Bit Budget=3 b/dim2026.0682.5
KIVIModel=Qwen2.5-14B, Bit Budget=3 b/dim2026.0682.4
KIVIModel=DS-R1-Qwen-7B, Bit Budget=3 b/dim2026.0682.3
KIVIModel=Qwen2.5-3B, Bit Budget=3 b/dim2026.0682.1
Block-GTQModel=DS-V2-Lite, Bit Budget=4 b/dim2026.0681.6
TQ-MSEModel=DS-R1-Llama-8B, Bit Budget=3 b/dim2026.0681.5
TQ-MSEModel=Llama-3.1-8B, Bit Budget=3 b/dim2026.0681.4
Block-GTQModel=DS-R1-Qwen-7B, Bit Budget=2 b/dim2026.0681.4
KIVIModel=DS-V2-Lite, Bit Budget=3 b/dim2026.0681.2
KIVIModel=Qwen3-8B, Bit Budget=4 b/dim2026.0681.1
KIVIModel=GLM-4-9B, Bit Budget=4 b/dim2026.0680.8
Block-GTQModel=Mistral-Nemo-12B, Bit Budget=2 b/dim2026.0680.5
TQ-MSEModel=GLM-4-9B, Bit Budget=4 b/dim2026.0679.6
TQ-MSEModel=DS-R1-Qwen-7B, Bit Budget=3 b/dim2026.0679.4
TQ-MSEModel=Qwen2.5-14B, Bit Budget=3 b/dim2026.0679.3
KIVIModel=Mistral-Nemo-12B, Bit Budget=4 b/dim2026.0678.8
KIVIModel=Qwen3-8B, Bit Budget=3 b/dim2026.0678.5
TQ-MSEModel=Qwen3-8B, Bit Budget=3 b/dim2026.0677.2
KIVIModel=Llama-3.1-8B, Bit Budget=2 b/dim2026.0677.2
Block-GTQModel=Llama-3.1-8B, Bit Budget=2 b/dim2026.0677.1
Block-GTQModel=DS-R1-Llama-8B, Bit Budget=2 b/dim2026.0677.1
KIVIModel=Mistral-Nemo-12B, Bit Budget=3 b/dim2026.0676.9
TQ-MSEModel=Qwen2.5-3B, Bit Budget=3 b/dim2026.0676.8
Block-GTQModel=Qwen3-8B, Bit Budget=2 b/dim2026.0676.3
KIVIModel=DS-R1-Llama-8B, Bit Budget=2 b/dim2026.0676.3
Block-GTQModel=GLM-4-9B, Bit Budget=3 b/dim2026.0676.1
KIVIModel=GLM-4-9B, Bit Budget=3 b/dim2026.0675.8
Block-GTQModel=Qwen2.5-14B, Bit Budget=2 b/dim2026.0675.7
Block-GTQModel=Qwen2.5-3B, Bit Budget=2 b/dim2026.0675.2
Block-GTQModel=Qwen3-30B, Bit Budget=2 b/dim2026.0674.4
TQ-MSEModel=DS-V2-Lite, Bit Budget=4 b/dim2026.0673.3
KIVIModel=DS-R1-Qwen-7B, Bit Budget=2 b/dim2026.0672.9
KIVIModel=Qwen3-30B, Bit Budget=2 b/dim2026.0672.6
TQ-MSEModel=Qwen3-30B, Bit Budget=3 b/dim2026.0672.5
TQ-MSEModel=Mistral-Nemo-12B, Bit Budget=2 b/dim2026.0672.5
KIVIModel=Qwen2.5-14B, Bit Budget=2 b/dim2026.0672.3
KIVIModel=Qwen2.5-3B, Bit Budget=2 b/dim2026.0671.6
KIVIModel=Mistral-Nemo-12B, Bit Budget=2 b/dim2026.0669.5
KIVIModel=Qwen3-8B, Bit Budget=2 b/dim2026.0669.4
Block-GTQModel=DS-V2-Lite, Bit Budget=3 b/dim2026.0668.9
TQ-MSEModel=Llama-3.1-8B, Bit Budget=2 b/dim2026.0668.3
TQ-MSEModel=DS-R1-Llama-8B, Bit Budget=2 b/dim2026.0668.2
KIVIModel=DS-V2-Lite, Bit Budget=2 b/dim2026.0668
TQ-MSEModel=DS-R1-Qwen-7B, Bit Budget=2 b/dim2026.0666.5
TQ-MSEModel=GLM-4-9B, Bit Budget=3 b/dim2026.0666
TQ-MSEModel=Qwen2.5-14B, Bit Budget=2 b/dim2026.0664.8
TQ-MSEModel=Qwen2.5-3B, Bit Budget=2 b/dim2026.0662.1
KIVIModel=GLM-4-9B, Bit Budget=2 b/dim2026.0661.7
TQ-MSEModel=Qwen3-8B, Bit Budget=2 b/dim2026.0661.6
Block-GTQModel=GLM-4-9B, Bit Budget=2 b/dim2026.0660.2
TQ-MSEModel=DS-V2-Lite, Bit Budget=3 b/dim2026.0655.4
TQ-MSEModel=Qwen3-30B, Bit Budget=2 b/dim2026.0655.2
Block-GTQModel=DS-V2-Lite, Bit Budget=2 b/dim2026.0649.9
TQ-MSEModel=GLM-4-9B, Bit Budget=2 b/dim2026.0647
TQ-MSEModel=DS-V2-Lite, Bit Budget=2 b/dim2026.0633.9