Asked about Attention, conceptually
What do multiple attention heads do?
An answer that works
Each head learns a different kind of relationship, such as grammatical agreement in one and topical relevance in another, and their outputs are combined. One set of weights can only represent one notion of relevance.
Reading an answer is not being able to give one. The section that teaches this is Attention, conceptually, 8 hours of Transformers & attention.
Learn this properlyAlso asked about Transformers & attention
Explain how attention works.Why did transformers replace recurrent networks?Why is long context expensive?What does causal masking do?What is the difference between attention and the feed-forward layer?Why does a transformer need position information at all?What are residual connections and layer norm for?What is the KV cache?
All 17 questions on Transformers & attention