Ajay Shenoy
← Back to NLP

Tokenization

Ajay Shenoy · July 2026 · placeholder

Before a model can process text, it has to be split into discrete units — tokens. This lesson is a placeholder for how modern tokenizers work.

What's coming

  • Word-level vs. subword vs. byte-level tokenization.
  • Byte-pair encoding, and why it's the default for most LLMs.
  • How tokenization choices affect context length and cost.