Repository files navigation

Torch and TorchVision, for your Node servers.Get up and running with PyTorch models within your NodeJS infrastructure in seconds.

---

Getting Started •Key Features •Development •Misc •License

Your models must be exported to torchscript in order to work with pytorchjs. Checkthis out for an example!

Assuming nothing's broken:yarn add pytorchjs

Run your PyTorch models in Javascript, just like you would in Python.

import{torch,torchvision}from'pytorchjs';const{ load}=torch;const{ DataLoader}=torch.utils.data;const{ ImageFolder}=torchvision.datasets;const{ Compose, Resize, InvertAxes, Normalize}=torchvision.transforms;constsqueezeNet=load("./test/resources/squeezenet_ts.pt");consttransforms=newCompose([newResize({height:224,width:224}),newNormalize([0.485,0.456,0.406],[0.229,0.224,0.225]),newInvertAxes()]);constloader=newDataLoader(newImageFolder("./test/resources/dataset"),1,transforms);constresults=awaitsqueezeNet(loader);

Additional examples of both setup and usage involving features like Torchvision Transforms and CUDA (in development) may be foundhere.

• Run your PyTorch models in a Javascript environment, without worrying about setting up Torchscript or downloading custom binaries
• Deploy your model using configurations identical to what you used during training
• Built-in CUDA support
  • CUDA support is a work in progress
• Support for TorchVision, including transforms, dataset classes, and pre-trained models
  • Support for TorchVision models is a work in progress

• yarn install should allow you to install project dependencies
• yarn test to run the test suite for this project

• This project usesarition's fork of torch-js to run TorchScript - check the project out if you're curious about how we do it!
• Distributed under the MIT license. SeeLICENSE for more information.
• This project was originally developed as a part ofCOMSW4995 - Open Source Development atColumbia University.