Fetching the paper…

Sigma: Differential Rescaling of Query, Key and Value for Efficient Language Models · Around