kgptalkie/mini-laya

Mini Laya

A small System One decision model, built from scratch in the course Fine-Tuning System One Decision Models. It has Laya's architecture on a ModernBERT-base encoder with a two-layer decision head, and it was trained with Laya's finetune on typed questions made from Kev's labelled eval sets.

Give it a state (a message, a ticket, an email) and typed questions; it returns typed answers with calibrated probabilities in one forward pass.

Use

import laya

agent = laya.Agent("kgptalkie/mini-laya")

Limits

Mini Laya is a teaching model. It is smaller than Laya and trained on far less data, so use Laya for real work.