froog 0.2.1

A SUPER SIMPLE MACHINE LEARNING FRAMEWORK

froog

FROOG: fast real-time optimization of gradients
a beautifully compact machine-learning library
homepage | documentation | pip

FROOG is a neural network framework that is actually SIMPLE with the goal of running machine learning on any device --> easily and efficiently.

Installation

pip install froog

Overview of Features

• Tensors
• Automatic Differentiation
  • Forward and backward passes
• Input/gradient shape-tracking
• MNIST example
• 2D Convolutions (im2col)
• Numerical gradient checking
• The most common optimizers (SGD, Adam, RMSProp)

Math Operations

• Scalar-Matrix Multiplication
• Dot Product
• Sum
• ReLU
• Log Softmax
• 2D Convolutions
• Avg & Max pooling
• More

Ready-to-Go Models

• EfficientNet-B0

Bounties

Want to help but don't know where to start? Here are some bounties for you to claim

Small

• binary cross entropy
• flatten
• batch_norm
• div
• pow
• dropout

Medium

• start doing ops with opencl
• einsum convs
• simplify how context and gradients are handled

Large

• ability training on FROOG!!!!
• float16 support
• transformers
• stable diffusion
• winograd convs
• GPU Support
  • MPS
  • CUDA
  • OpenCL

Contributing

Here are the rules for contributing:

• increase simplicity
• increase efficiency
• increase functionality

more info on contributing