PyTorch Serving Workshop

Overview

This repository contains notebooks for a PyTorch model-serving workshop.

You do not need a GPU runtime.

Setup

During the workshop, use this custom JupyterHub, which has all dependencies preinstalled.

Outside the workshop, use .

Contents

There are five notebooks.

a. 00_prepare_dataset.ipynb

Prepares and saves the e-commerce dataset.

b. 01_train.ipynb

Trains a DistilBERT model.

c. 02_inference_review.ipynb

Introduces the Hugging Face ecosystem and shows how to use the trained model from the previous notebook.

d. 03_optimizing_model.ipynb

Demonstrates the impact of quantization and TorchScript.

e. 04_packaging.ipynb

Shows how to package and serve models with TorchServe.

Slides

Video

References

Pydata 2021 Slides

Pydata 2021 Conference Page

Libraries

This repository uses the Hugging Face Transformers and Datasets packages.

The dataset used is Amazon Berkeley Objects (ABO) Dataset created by Amazon and UC Berkeley. For more information, see the accompanying paper.

Contact

For help or feedback, please reach out to:

• Nidhin Pattaniyil
• Adway Dhillon
• Vishal Rathi