← All questions
HardCodingSystem design

Implement a Transformer block from scratch

1Give yourself 5 minutes
2Answer out loud, not in your head
3Then compare with the answer below

Reference answer

Then expect these follow-ups

  • Add positional information — compare learned embeddings, sinusoidal, and RoPE; why RoPE won for long context. Parameter count of one block as a function of d_model?

  • (~12·d_model² with d_ff = 4·d_model.) Decoder-only vs encoder-decoder — what changes structurally?

  • Where does KV caching plug into your `forward` at inference time?

Free to read · better with Enzo

Practice this out loud with Enzo

Enzo runs it as a mock interview, pushes back with follow-ups, and grades you on the rubric.

Next question