News for hackers. Stuff that matters.

A comment on Our Humble Attempt at “How Much Data Is Needed to Fine-Tune. Parent comment

Re:Our Humble Attempt at “How Much Data Is Needed to Fine-Tune by KRAKRISMOTT on Sunday September 24, 2023 @10:36PM (#37639103)
Why not use the finite state machine logits filter trick for controlling output format? It seems to work pretty well for JSON
[ Parent | on HN ]
Re:Our Humble Attempt at “How Much Data Is Needed to Fine-Tune by antman on Monday September 25, 2023 @01:29AM (#37639943)
What is that? Sounds useful
[ Parent | on HN ]
Re:Our Humble Attempt at “How Much Data Is Needed to Fine-Tune by evrimoztamur on Monday September 25, 2023 @04:01AM (#37640772)
I think he's referring to https://news.ycombinator.com/item?id=37125118

Controlling token outputs to match a regex generates valid JSON 100% of the time!

[ Parent | on HN ]