List your company

Long-context language understanding on RULER (4k Context)

0FWE RateE2LLM-R
-2.973217.095937.16557.2341Sep 10, 2024Dec 20, 2024Apr 1, 2025Jul 11, 2025Oct 21, 2025Jan 30, 2026May 12, 2026

Evaluation results

MethodLinks
E2LLM-RContext Length=4K2024.090————3———8.281.535
E2LLM-C+RContext Length=4K2024.092————7.2———66.2232.437
LongLLMLinguaContext Length=4K2024.094.67————5.2———27.447.614.5
LongLoRAContext Length=4K2024.099.33————1.6———41.0716.655.5
Full AttentionAttention mechanism=Full Attention, Budget=Full context, Backbone=Llama3-8B-Instruct-Gradient-1048k2026.021010089.687.879.217.479.856.463.7———
LycheeDecodeAttention mechanism=Sparse Attention, Budget=4096 tokens, Backbone=Llama3-8B-Instruct-Gradient-1048k2026.021010089.488.478.917.38056.263.7———
E2LLM-CContext Length=4K2024.0913.67————1.2———2.456.835.5
YaRNContext Length=4K2024.0920.33————9.8———4415.27
LLoCOContext Length=4K2024.0924.67————0———0.4427.732.5
CCA-attentionBackbone=LLaMA2-7B-32K2026.0545.67—————————41.1—
StreamingLLMBackbone=LLaMA2-7B-32K2026.0547.83—————————61.8—
LLaMA2-7B-32K (Vanilla Self-Attention)Backbone=LLaMA2-7B-32K, Attention=Vanilla Self-Attention2026.0548—————————60.65—
MInferenceBackbone=LLaMA2-7B-32K2026.0548.67—————————60.35—
DuoAttentionBackbone=LLaMA2-7B-32K2026.0548.83—————————60.15—
Training–Inference Consistent Segmented ExecutionBackbone=LLaMA2-7B-32K2026.0553.83—————————70.6—
LLama2-7B-chatContext Length=4K2024.0974.33————27———95.8785.663