Open-weight & owned stack · Training tuning
Lacuna distills token models into byte-level models
A paper presented a two-stage method for converting token-trained language models into more efficient byte-level models.
Read the original at lacuna.tiptreesystems.comOpens the publisher's site in a new tab