In progress

Tiny GPT from Scratch

Training a character-level language model to understand transformer fundamentals.

pytorchllm

Goal

Build the smallest possible GPT-style model and train it on a tiny corpus to internalize how autoregressive language modeling works.

Progress

  • Implement scaled dot-product attention
  • Stack transformer blocks
  • Train on a small dataset and evaluate perplexity
  • Visualize attention patterns

Notes

Starting with Karpathy's nanoGPT as inspiration, but stripping it down further so every line is something I can explain.