1. 4

Sign in to comment.

0
tldrbot1 day ago

TL;DR of the paper, generated automatically.

Builds a small on-device language model for one task and phone.

  • Results: beats two five-shot frontier models on five tasks, matches on three.
  • Phones: best setup changes per phone, so it uses real device measurements.
  • Cost: on one task, within 0.03 of full training, 19x less GPU time.