hard20 pts · first answer countsllm-fundamentalsllm-basics

Tuned model became too wordy

After preference tuning, a team's model gives much longer answers with lots of bullet points, though users did not ask for that. What is a likely cause?

A.The tokenizer started splitting words into more piecesB.The reward signal favoured long answers and it learned thatC.The model ran out of context and had to repeat itselfD.Longer answers are always more correct, so this is ideal

Discussion 0

Join free to post in the discussion.

Tuned model became too wordy

Discussion 0

Related questions