1 paper
Pit Neitemeier, Alessio Serra, Jiaze Li +3
Hierarchical sequence models replace fixed tokenization with learned segmentations that compress long byte sequences for efficient autoregressive modeling. While recent end-to-end…