List your company

Delta Accuracy on RULER Long-context Understanding

6.9Delta AccuracyCertified Quantized Attention
0.141.8953.655.405May 20, 2026

Evaluation results

MethodLinks
Certified Quantized AttentionContext Length=8K, n=20 slices, Backbone=LLaMA 3.1-8B, Key Precision=INT8 (asymmetric), Value Precision=INT4 (g=16)2026.056.9
Certified Quantized AttentionContext Length=32K, n=20 slices, Backbone=LLaMA 3.1-8B, Key Precision=INT8 (asymmetric), Value Precision=INT4 (g=16)2026.051.9
Certified Quantized AttentionContext Length=64K, n=20 slices, Backbone=LLaMA 3.1-8B, Key Precision=INT8 (asymmetric), Value Precision=INT4 (g=16)2026.051.5
Certified Quantized AttentionContext Length=128K, n=20 slices, Backbone=LLaMA 3.1-8B, Key Precision=INT8 (asymmetric), Value Precision=INT4 (g=16)2026.050.4