不同的计算K的策略的耗时

[ywlq]

论文中的表6显示了如何去计算当前step的k
我的理解是只需要保存下过去的k然后进行加权平均，只需要少量的矩阵运算，为什么会这么耗时，例如表中的k_avg相对于k_local来说延迟增加了38s

[LMD0311]

Thank you for your interest in our work.

We apologize for any misunderstanding this may have caused. This table explores different methods for updating k, which affects cache utilization.

For example, k_avg considers the average of all historical k values, while k_ema gradually reduces the weight of older k values. We found that using the current k value (k_local) directly achieves a favorable trade-off. Calculating k itself incurs negligible computational overhead, but it influences cache decisions.