List your company

Multiple Choice Question Answering on LongBench v2 (test)

44.07Accuracy (Easy, Short)Dense
38.776440.150741.52542.8993Jun 3, 2025

Evaluation results

MethodLinks
DenseBackbone=Llama3.1, Attention Type=Full Dense Attention2025.0644.0728.4131.1132.2325.9825.430.42
top-kBackbone=Llama3.1, sparse token budget=1024, Attention Type=Exact top-k attention2025.0640.682533.3329.7525.223.8128.63
HATABackbone=Llama3.1, sparse token budget=10242025.0638.9827.2735.5629.7526.7725.429.62