GenAI Platform Glossary
DigitalOcean GenAI Platform is an offering for building AI agents on GPU-powered infrastructure. Use the platform to create AI applications with foundation models and function and agent routes, knowledge bases, and Retrieval-Augmented Generation (RAG) pipelines. See Features for more information.
This glossary defines the core concepts behind the GenAI Platform to help build your mental model of how the GenAI platform works and understand what the documentation is referring to when it uses certain terminology.
Agent instructions are a plain language model setting that establish to your agent what its objective is, how it should behave, and any other rules you want it to follow.
Generative Artificial Intelligence (AI) are machine learning models that use different types of transformer or diffusion-based models to analyze knowledge bases and, based on a large corpus of data, can mimic its patterns to produce new content, such as text, images, and music.
K-Value is a semantic search setting that determines how many objects your model can return from knowledge base queries.
Max tokens is a model setting that determines the maximum number of tokens that the model can generate in its responses. It ranges from 0 to 512, with 256 as a default.
Retrieval Augmented Generation (RAG) is a technique that enhances a model’s responses by integrating information from a Knowledge Base (KB). By referencing a KB, the model can generate responses that include data not originally part of its training, leading to more accurate and contextually relevant answers.
Temperature is a model setting that determines how creative a model’s response is. Much like the Top P setting, it determines diversity. A lower temperature instructs the model to deviate less from its data, while a higher temperature allows the model to extrapolate more information and deviate from its data.
A token is the basic unit of text or data that a foundational model processes. It can represent a word, part of a word, or even a single character.
Top P is a model setting that determines how many different phrases a model considers when it’s trying to predict the next string in an output.
A vector database is a specialized database for storing, indexing, and querying high-dimensional data, commonly used in machine learning (ML) and artificial intelligence (AI). High-dimensional data represent objects like images or text in mathematical form, using vectors.
Vector embedding is a way to represent data as a matrix of numbers in a high-dimensional space. This allows your model to understand the data, its meaning, and the relationships between different pieces of data.