1. 1

Sign in to comment.

0
tldrbot5 hours ago

TL;DR of the paper, generated automatically.

A new dataset and fine-tuned model for Even, an endangered language spoken in Siberia.

  • Data: 50,521 Even-Russian entries, expanded to 107,543 training examples.
  • Tokenizer: 27% fewer tokens per word, down from 4.99 to 3.64.
  • VQA: 65% accuracy, up from 0% for the base model.