Overview

There are a number of different techniques available in Predibase via the Ludwig framework for building and customizing LLMs, including:

• Zero Shot Learning for using LLMs to solve tasks and answer questions when no ground truth data is available.
• Few Shot Learning for using LLMs to solve tasks by giving it a handful of examples to help guide it.
• Fine-Tuning for modifying the weights of the LLM to adapt it to new data and tasks.

The flowchart below visually presents a simplfiied decision-making process that can be used to decide which customization technique is right for you: