https://giuliapl.github.io/tags/transformers/Recent content in Transformers on giuliapl personal page & blogHugo -- 0.156.0en-enFri, 15 May 2026 00:00:00 +0000https://giuliapl.github.io/posts/attention-in-transformers/Fri, 15 May 2026 00:00:00 +0000https://giuliapl.github.io/posts/attention-in-transformers/Implementing a single self-attention head in ~30 lines of NumPy. What surfaces when you write the math by hand: the role of √d_k, where the causal mask goes, and why the row/column convention is the thing that actually trips you up.