Training a small model on our data: from 21 to 90 out of 98
We trained Laya, a small free model, to grade answers: from 21 to 90 of 98 cases passed after 10 minutes of training. The numbers and the limits.
· 4 min readRead the article
Studies run on our own products and our own data, with the numbers, the method and the limits. No generic leaderboard: what worked for us, and what didn't.
We trained Laya, a small free model, to grade answers: from 21 to 90 of 98 cases passed after 10 minutes of training. The numbers and the limits.
Four AI models tested on our own data: which one answers best, which one grades best, and the cost cut we reverted.
Planning a cloud migration? Looking to integrate AI into your processes? Need an expert review of your architecture? Let's talk.