Sparse Operation Kit Notebooks

This directory contains a set of Jupyter notebooks that demonstrate how to use SOK in TensorFlow.

Quickstart

The simplest way to run a one of our notebooks is with a Docker container. A container provides a self-contained, isolated, and reproducible environment for repetitive experiments. Docker images are available from the NVIDIA GPU Cloud (NGC). If you prefer to build the HugeCTR Docker image on your own, refer to Set Up the Development Environment With Merlin Containers For SOK.

Pull the NGC Docker

Pull the container using the following command:

docker run nvcr.io/nvidia/merlin/merlin-tensorflow:nightly

Clone the HugeCTR Repository

Use the following command to clone the HugeCTR repository:

git clone https://github.com/NVIDIA/HugeCTR

Start the Jupyter Notebook

Launch the container in interactive mode (mount the HugeCTR root directory into the container for your convenience) by running this command:

docker run --runtime=nvidia --rm -it --cap-add SYS_NICE -u $(id -u):$(id -g) -v $(pwd):/hugectr -w /hugectr -p 8080:8080 nvcr.io/nvidia/merlin/merlin-tensorflow:23.11

Start Jupyter using these commands:

cd /hugectr/hps_tf/notebooks
jupyter-notebook --allow-root --ip 0.0.0.0 --port 8080 --NotebookApp.token='hugectr'

Connect to your host machine using the 8080 port by accessing its IP address or name from your web browser: http://[host machine]:8080

Use the token available from the output by running the command above to log in. For example:

http://[host machine]:8080/?token=aae96ae9387cd28151868fee318c3b3581a2d794f3b25c6b

Notebook List

Here’s a list of notebooks that you can run:

sok_train_dlrm_demo.ipynb: Demonstrates how to train DLRM model with TF and SOK.
hps_inference_dlrm_with_sok_weight_demo.ipynb: Demonstrateshow to use hps-tf to read SOK’s sparse weight and perform inference. Before running, you must first complete the training in sok_train_dlrm_demo.ipynb.
sok_incremental_dump_demo.ipynb: Demonstrates how to use SOK for incremental dump and verifies its correctness.
sok_dump_load_demo.ipynb: Demonstrates how to use SOK for dump/load and verifies the correctness of the dump/load operations.

System Specifications

The specifications of the system on which each notebook can run successfully are summarized in the table. The notebooks are verified on the system below but it does not mean the minimum requirements.

Notebook	CPU	GPU	#GPUs	Author
sok_train_dlrm_demo.ipynb	AMD EPYC 7742 64-Core Processor 2 TB Memory	Ampere A100-SXM4-80GB 80 GB Memory	2	Hui Kang
hps_inference_dlrm_with_sok_weight_demo.ipynb	AMD EPYC 7742 64-Core Processor 2 TB Memory	Ampere A100-SXM4-80GB 80 GB Memory	1	Hui Kang
sok_incremental_dump_demo.ipynb	AMD EPYC 7742 64-Core Processor 2 TB Memory	Ampere A100-SXM4-80GB 80 GB Memory	2	Hui Kang
sok_dump_load_demo.ipynb	AMD EPYC 7742 64-Core Processor 2 TB Memory	Ampere A100-SXM4-80GB 80 GB Memory	2	Hui Kang