List your company

Long-context evaluation on LongBench (test)

32.85NarQA ScorexKV
6.985213.700120.41527.1299Mar 10, 2025

Evaluation results

MethodLinks
xKVBase Model=Llama-3.1-8B-Instruct, K.V. BW (x ↓)=8.032025.0332.8545.6254.6951.7436.3828.3331.3224.4922.6126.2610041.3
PyramidKVBase Model=Llama-3.1-8B-Instruct, K.V. BW (x ↓)=8.002025.0331.8344.3652.5656.3848.3527.2328.6325.0623.4924.899.542.02
TokenButlerBase Model=Llama-3.1-8B-Instruct, K.V. BW (x ↓)=8.002025.0330.7348.8656.4557.7750.8228.3434.3425.326.9129.2210044.43
DenseBase Model=Llama-3.1-8B-Instruct, K.V. BW (x ↓)=1.002025.0330.6446.6455.4258.1747.7628.2334.5125.6125.4323.5110043.27
SnapKVBase Model=Llama-3.1-8B-Instruct, K.V. BW (x ↓)=8.002025.0330.3445.5653.6257.9247.526.9628.9925.3323.9524.9199.542.23
PyramidKVBase Model=Qwen2.5-7B-Instruct-1M, K.V. BW (x ↓)=8.002025.0329.9341.2947.2160.1651.9832.7927.2122.8819.864510043.48
KIVIBase Model=Llama-3.1-8B-Instruct, K.V. BW (x ↓)=7.102025.0329.8441.4951.2857.2742.5427.5433.2825.0123.925.8898.1741.47
TokenButlerBase Model=Qwen2.5-7B-Instruct-1M, K.V. BW (x ↓)=8.002025.0329.3947.2752.559.6756.0134.5833.3223.1424.0435.0910045
DenseBase Model=Qwen2.5-7B-Instruct-1M, K.V. BW (x ↓)=1.002025.0329.2143.7848.5860.9253.2933.6833.2323.2423.5343.2110044.79
xKVBase Model=Qwen2.5-7B-Instruct-1M, K.V. BW (x ↓)=8.032025.0328.7842.9847.4358.7952.4232.6933.0223.6723.1838.629843.6
SnapKVBase Model=Qwen2.5-7B-Instruct-1M, K.V. BW (x ↓)=8.002025.0328.6844.5147.4360.8851.9132.1729.9523.0521.7945.6710044.19
Single SVDBase Model=Qwen2.5-7B-Instruct-1M, K.V. BW (x ↓)=8.402025.0327.8841.8447.6652.7348.7628.4628.7122.2421.1833.186537.97
KIVIBase Model=Qwen2.5-7B-Instruct-1M, K.V. BW (x ↓)=7.102025.0326.4335.7838.446.1745.0823.7924.6222.4116.9839.5263.534.79
StreamingLLMBase Model=Llama-3.1-8B-Instruct, K.V. BW (x ↓)=8.002025.0326.2529.3234.5649.9942.4720.9226.2921.922.6323.8793.535.61
StreamingLLMBase Model=Qwen2.5-7B-Instruct-1M, K.V. BW (x ↓)=8.002025.0324.1334.9927.4747.5545.1321.4426.7619.3421.1544.112931.01
MiniCacheBase Model=Llama-3.1-8B-Instruct, K.V. BW (x ↓)=1.302025.0318.1120.4127.5138.821.3720.1921.0920.8519.622.759829.88
Single SVDBase Model=Llama-3.1-8B-Instruct, K.V. BW (x ↓)=8.402025.0315.5238.3637.1430.0419.722.0619.5222.3518.9122.828530.13
MiniCacheBase Model=Qwen2.5-7B-Instruct-1M, K.V. BW (x ↓)=1.302025.037.9816.479.4221.7815.4610.558.837.943.8411.32510.78