"A query regarding the acquisition of v values."

[skipsupper]

Thank you and your team for your work; this is excellent research. However, I have a question to ask: In the original model, the queries (q), keys (k), and values (v) all pass through linear layers, but in your newly designed model, the v values are obtained without going through a linear layer. Why was this design choice made? Are there any related ablation experiments? Thank you for your reply.