LoGo: Token-Level Dynamic Local-Global Attention
尚未评估计划受阻指标结果claim-short-context-recall
On the paper's 32k-checkpoint short-context recall aggregate, LoGo is reported as competitive with the other methods and scores 57.65 percentage points across six recall tasks.
来源：paper-fixed:Section 4.2 and Table 3, p. 7
报告指标与观测值
short_context_recall_average
m-table3-logo-recall
论文报告 57.65 percentage_points
实际观测 — percentage_points
Assessment（0）
这条 Claim 暂无已发布的不可变 Assessment。
关联运行（0）
这条 Claim 暂无关联执行。