The Many Minds of Transformers

Medium4 min read

What each attention head is really learning, and why running several in parallel gives models richer representations.

Originally published on

Medium

Read the full article