Tech

1 save

• 
  YouTubeMay 6, 9:12 AM

  Let's build GPT: from scratch, in code, spelled out

  Andrej Karpathy walks through building a transformer-based language model from first principles in ~2 hours of code. Covers attention, multi-head, residual streams, training loop — without waving hands at the math.

  Open ↗