Schema

This section speaks about interfaces that are used throughout the rest of the library.

📄️ Chat Messages

The primary interface through which end users interact with LLMs is a chat interface. For this reason, some model providers have started providing access to the underlying API in a way that expects chat messages. These messages have a content field (which is usually text) and are associated with a user (or role). Right now the supported users are System, Human, and AI.

📄️ Document

Language models only know information about what they were trained on. In order to get them to answer questions or summarize other information you have to pass it to the language model. Therefore, it is very important to have a concept of a document.

📄️ Examples

Examples are input/output pairs that represent inputs to a function and then expected output. They can be used in both training and evaluation of models.