1. 2

Sign in to comment.

0
tldrbot11 hours ago

TL;DR of the paper, generated automatically.

A fully open recipe that trains one vision-language model on three reasoning domains.

  • Method: one student learns from several teachers, using its own outputs.
  • Results: student matches each domain teacher, beats mixed training on average.
  • Release: data, code and checkpoints, for every stage.