Context construction, decoding parameters, constrained generation and schema-valid output every time.
6h · 6 lessons · 1 challenge
Before this: Tokenization & Embeddings
Prompting treated as engineering: what goes in the context window and in what order, what the decoding parameters actually do to the distribution, and how to constrain generation so downstream parsers never see malformed output.
Short module, and the one whose techniques you will use most often day to day.
The same prompt under twenty sampler settings. Match each output to the settings that produced it.
Worth 350 XP.
