PyTorch Perforated Backpropagation installs statistics

Are you building a PyTorch project where you would benefit from up to 40% improved accuracy or 90% reduction in model size without a loss in accuracy? Perforated Backpropagation is a new optimization tool developed by Perforated AI that allows these improvements with only minutes of changes to your code.

Perforated Backpropagation is a PyTorch optimization tool that can be added to existing pipelines with only a few lines of code. Almost everything is done behind the scenes with only minor edits to your python scrips. Integrating our app provides the installation of the package granting access to the functions required to use this novel technique. After installation the two main functions which need to be added are just converting your model and passing a tracker object information about when to run optimization cycles.

This block belongs at the beginning of your process:

# Define Your Model model = yourModel() # Convert your modules to be Perforated Backpropagation Modules model = PBU.convertNetwork(model)

And the following goes in your validation loop:

optimizedModel = PBG.pbTracker.addValidationScore(score)

Additional details about implementation can be found in our documentation.

Description from the GitHub Marketplace listing.