Build your own GPT-2
Multi-head attention·2 min

Why multiple heads

0 / 1

Before assembling all the heads, one conceptual question: why split attention into 12 heads at all?

Knowledge check

Why split attention into 12 heads instead of one big 768-wide head?