Q, K, V & Causal Attention During Inference

What queries, keys and values do in attention, why generation needs a causal mask, and what a real attention pattern captured from Llama 3.1 8B looks like.