Overview

At Maxim we have a robust datasets capability for evaluation, in-context learning and fine-tuning. Datasets are very critical for AI application development, but creating, updating, and managing datasets entails a lot of manual effort and is a big overhead for development teams. Maxim’s data management engine makes it simple for you and your teams to easily create, update, and manage datasets throughout your development lifecycle.

Managing your datasets on Maxim consists of 3 parts:

1. Creating datasets - Create from scratch or upload CSV
2. Using datasets during evaluation
3. Continuous curation of datasets from test runs, human annotations or production logs

Datasets provides all the functionality of spreadsheets with more intelligence and easy connection to the overall evaluation pipelines.

• Flexible like excel , bring or create any number of columns
• Easy creation, deletion, updating, and duplication features
• Ability to import CSV

On Maxim, you also have the ability to create splits for different logical groups and use these while testing.

While evaluating a prompt or workflow, a dataset can be chosen. Inputs and variables are then replaced and used to generate the outputs for all entries. If you already have the outputs, they can be added in an output type column of the dataset and the dataset can be evaluated independently.