Causal Masking

In an autoregressive Transformer decoder, causal masking forces every attention score for a future position to negative infinity before the softmax, so token…

This lesson is part of BitBootcamp Pro. Unlock it with Pro, or browse the free courses.