https://github.com/project-codeflare/zero-copy-model-loading

In-depth code associated with my Medium blog post, "How to Load PyTorch Models 340 Times Faster with Ray"
https://github.com/project-codeflare/zero-copy-model-loading

Last synced: about 1 month ago
JSON representation

In-depth code associated with my Medium blog post, "How to Load PyTorch Models 340 Times Faster with Ray"

• Host: GitHub
• URL: https://github.com/project-codeflare/zero-copy-model-loading
• Owner: project-codeflare
• License: apache-2.0
• Created: 2021-09-20T17:14:12.000Z (over 3 years ago)
• Default Branch: main
• Last Pushed: 2022-09-02T22:22:12.000Z (over 2 years ago)
• Last Synced: 2023-08-29T03:45:00.383Z (over 1 year ago)
• Language: Jupyter Notebook
• Homepage: https://medium.com/ibm-data-ai/how-to-load-pytorch-models-340-times-faster-with-ray-8be751a6944c
• Size: 3.55 MB
• Stars: 17
• Watchers: 1
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# zero-copy-model-loading

Code to accompany the Medium post, ["How to Load PyTorch Models 340 Times Faster

with

Ray"](https://medium.com/ibm-data-ai/how-to-load-pytorch-models-340-times-faster-with-ray-8be751a6944c).

## Notebooks

Notebooks can be found in the `notebooks` directory:

* `zero_copy_loading.ipynb`: The notebook that was used when authoring the 

  original blog post.

* `benchmark/benchmark.ipynb`: The notebook that was used when authoring the

  second post in the series.

Instructions to run notebooks:

1. Install `bash` and either `anaconda` or `miniconda`.

1. Check out a copy of this repository and navigate to the root directory of

   your local copy.

1. Run the script `./env.sh`, which creates an Anaconda environment under 

   `/env`.

   ```

   ./env.sh

   ```

1. Activate the Anaconda environment:

   ```

   conda activate ./env

   ```

1. Run JupyterLab:

   ```

   jupyter lab

   ```

1. Navigate to the `notebooks` directory and open up the Jupyter notebook of your choice.

## Python Package

This repository also contains the source code for the `zerocopy` library.

`zerocopy` is a Python package that provides functions for implementing

zero-copy model loading of PyTorch models on Ray.

You can find the source code for the package inside the `package` directory.