Introducing TorchExpo

Collection of models and extensions for mobile deployment in PyTorch

UPDATE Oct 02, 2020:
TorchExpo won 3rd place @ PyTorch Global Summer Hackathon 2020, read more here.


What is TorchExpo?

TorchExpo is an initiative which consists of python package and mobile app + website. If you are a ML Developer that has worked with PyTorch and you want to go from research to a mobile solution, TorchExpo might help you. If you just want to incorporate some pre-trained models in your mobile devices without any hassle, TorchExpo got lot of solutions in different formats!

It is a collection of models and extensions for mobile deployment built on PyTorch ecosystem. We all know PyTorch is cool. You don’t know? Check out this Papers With Code Trends. With TorchExpo, you get to convert a SoTA model into mobile-friendly format in just few lines. If you don’t want to convert yourself, you can always download them via the website. This project is inspired by PyTorch Forum’s #Mobile Category and Tensorflow Hub.

Why it might be useful?

And finally, wouldn’t it be 100x better that you don’t have to spend time converting PyTorch model in different formats just to deploy on a mobile device?

Show Me An Example!

Image Classification

from torchexpo.vision import image_classification

model = image_classification.squeezenet1_0()
model.extract_torchscript()
model.extract_onnx()

Image Segmentation

from torchexpo.vision import image_segmentation

model = image_classification.fcn_resnet50()
model.extract_torchscript()
model.extract_onnx()

Sentiment Analysis

from torchexpo.vision import sentiment_analysis

model = image_classification.electra_imdb()
model.extract_torchscript()
model.extract_onnx()

And you my friend, have converted few models in two mobile-friendly formats! Read more about other models at TorchExpo Docs.

Roadmap

Its time for sown seeds to grow with proper care such as:

I am becoming seriously allergic to 500-pound websites.