Topic: Generating text

Part of NLP and argumentation.

This topic contains 4 articles.

• Constrained decoding — Forcing a language model’s output to follow a grammar, so that it always parses.
• In-context learning — Teaching a language model a task by putting solved examples in its prompt, without training.
• LoRA fine-tuning — Adapting a model to a task by training small add-on matrices while its original weights stay frozen.
• Prompt caching — Reusing a language model’s work on a shared prompt prefix across many calls.