List your company

Long-context language understanding on RULER 64k context length

0FWE (Error)Full Attention
-3.652821.003645.6670.3164Sep 10, 2024Dec 20, 2024Apr 1, 2025Jul 11, 2025Oct 21, 2025Jan 30, 2026May 12, 2026

Evaluation results

MethodLinks
Full AttentionAttention mechanism=Full Attention, Budget=Full context, Backbone=Llama3-8B-Instruct-Gradient-1048k2026.02099.498.496.893.719.870.447.665.8———————
LycheeDecodeAttention mechanism=Sparse Attention, Budget=4096 tokens, Backbone=Llama3-8B-Instruct-Gradient-1048k2026.02099.673.290.381.719.963.444.459———————
E2LLM-CContext Length=64K2024.096.33————0.6————29.52.634.9———
E2LLM-RContext Length=64K2024.0910.67————4.4————150.132.6———
LongLLMLinguaContext Length=64K2024.0913.33————0.2————1523.250.3———
LLoCOContext Length=64K2024.0915.67————0————90.312.4———
MiniCacheBackbone=Qwen2.5-14B-Instruct-1M, Type=Inter, Compression Rate=1.30, Context Length=64K2025.0318.435.42—1.561.821.4620.8330.2113.78————12.513.542.08
E2LLM-C+RContext Length=64K2024.0932.67————6.2————34.59.340.2———
Training–Inference Consistent Segmented ExecutionBackbone=LLaMA2-7B-32K2026.0534.17———————————2———
MiniCacheBackbone=Llama-3.1-8B-Instruct, Type=Inter, Compression Rate=1.30, Context Length=64K2025.0358.3389.58—21.3514.0649.3861.4635.4245.04————66.6743.7510.42
Single SVDBackbone=Llama-3.1-8B-Instruct, Type=Intra, Compression Rate=8.40, Context Length=64K2025.0361.1125—44.7928.913.5447.9236.4645.71————51.0461.4696.88
PyramidKVBackbone=Llama-3.1-8B-Instruct, Type=Intra, Compression Rate=8.00, Context Length=64K2025.0368.06100—98.4410095.4283.3357.2989.94————10010096.88
SnapKVBackbone=Llama-3.1-8B-Instruct, Type=Intra, Compression Rate=8.00, Context Length=64K2025.0368.75100—97.661009583.3358.3389.68————10098.9694.79
xKVBackbone=Llama-3.1-8B-Instruct, Type=Inter, Compression Rate=8.03, Context Length=64K2025.0378.47100—96.6296.0986.6778.1356.2588.5————96.8897.9297.92
Single SVDBackbone=Qwen2.5-14B-Instruct-1M, Type=Intra, Compression Rate=6.35, Context Length=64K2025.0380.2194.79—66.1582.8146.0448.965071.79————66.6783.3398.96
KIVI-2Backbone=Llama-3.1-8B-Instruct, Type=Intra, Compression Rate=7.10, Context Length=64K2025.0384.38100—89.5891.4181.4680.2155.2186.87————96.8898.9690.63
Full AttnBackbone=Llama-3.1-8B-Instruct, Type=None, Compression Rate=1.00, Context Length=64K2025.0385.42100—97.6698.9697.2983.3359.3891.89————10098.9697.92
xKV-4Backbone=Qwen2.5-14B-Instruct-1M, Type=Inter, Compression Rate=6.21, Context Length=64K2025.0385.76100—91.931009571.8860.4290.19————97.9210098.96
PyramidKVBackbone=Qwen2.5-14B-Instruct-1M, Type=Intra, Compression Rate=6.00, Context Length=64K2025.0386.11100—97.1499.7499.5883.3364.5891.17————10010081.25
KIVI-2Backbone=Qwen2.5-14B-Instruct-1M, Type=Intra, Compression Rate=6.40, Context Length=64K2025.0389.58100—92.9798.1897.7178.1364.5890.45————98.9695.8388.54
SnapKVBackbone=Qwen2.5-14B-Instruct-1M, Type=Intra, Compression Rate=6.00, Context Length=64K2025.0390.28100—95.5799.7499.7981.2565.6391.66————10010084.38
Full AttnBackbone=Qwen2.5-14B-Instruct-1M, Type=None, Compression Rate=1.00, Context Length=64K2025.0391.32100—98.9610099.5880.2164.5893.36————10010098.96
CLAAModel=LLaMA-3.1-8B, Context length=64K, Token keep rate=40%2026.02——87.7286.67————82.0189.8563.8—————
FastKVModel=LLaMA-3.1-8B, Context length=64K, Token keep rate=40%2026.02——87.1686.67————81.2887.6963.6—————
FullKVModel=LLaMA-3.1-8B, Context length=64K, Token keep rate=40%2026.02——86.8886.67————83.7898.2763.3—————
GemFilterModel=LLaMA-3.1-8B, Context length=64K, Token keep rate=40%2026.02——58.288.33————73.1282.2363.7—————