Open-weight & owned stack · Training tuning
PITA aligns language models without additional training
PITA proposed preference-guided inference-time alignment to adapt language-model outputs without further model training.
Read the original at lacuna.tiptreesystems.comOpens the publisher's site in a new tab