Scaling Large Datasets with Criteo

Criteo provides the largest publicly available dataset for recommender systems with a size of 1TB of uncompressed click logs that contain 4 billion examples. We demonstrate how to scale NVTabular, as well as:

• Use multiple GPUs and nodes with NVTabular for feature engineering.

• Train recommender system models with the NVTabular dataloader for PyTorch.

• Train recommender system models with the NVTabular dataloader for TensorFlow

• Train recommender system models with HugeCTR using multiple GPUs.

• Inference with the Triton Inference Server and TensorFlow or HugeCTR.

There are example compose files for use with Docker in the scaling-criteo <https://github.com/NVIDIA-Merlin/Merlin/tree/main/examples/scaling-criteo>_ directory of the Merlin repository on GitHub. The compose files enable you to run a pair of training and inference containers.

Explore the following notebooks:

• Download and Convert

• Feature Engineering with NVTabular

• Training with FastAI

• Training with HugeCTR

• Training with TensorFlow

• Deploy the HugeCTR Model with Triton Inference Server

• Deploy the TensorFlow Model with Triton Inference Server