Bringing Your Own Dataset

This guide explains how to import your own dataset to be used with NeMo Automodel.

Types of Supported Datasets

NeMo Automodel supports several types of datasets for different training scenarios:

• Completion datasets: Single text sequences for language modeling

• Conversation datasets: Multi-turn chat dialogues

• Instruction datasets: Question-answer pairs

• Multi-modal datasets: Text combined with images and audio

TODO: onboard CPT and multi-turn SFT. Add documentation for all of the above.