Apache TVM is an open-source ML compiler framework for FPGAs, microcontrollers, browsers, CPUs, and GPUs. It works with popular frameworks like PyTorch and TensorFlow. The product was developed by OctoML to help engineers run computations efficiently on any hardware backend. 

Project Background

• Framework: Apache TVM
• Author: OctoML
• Released: 2017
• Type: Open-source ML compiler framework for CPUs, GPUs, and accelerators 
• License: Apache 2.0
• Supports: Works with PyTorch, TensorFlow, Keras, MXNet, and others
• Language: C++ and Python
• GitHub: apache/tvm with 7k contributors and 591 contributors

Features

• A universal compiler that ingests data from different frameworks and enables them to work on different devices from CPUs to GPUs to mobile devices to browsers 
• End to end deep learning compilation stack
• Problem it addresses: 1) Portability-allows users to deploy on different devices 2) Efficiency – squeezes as much performance as possible from hardware 3) Support-built a stack that works with hardware