Positional Encoding

XLNet Relative Positional Encoding

Sophisticated mechanism in XLNet that models relative distances between tokens in attention computation, enabling better generalization across different sequence lengths.

← Kembali