Apache TVM, an end-to-end machine compiler framework for CPUs, GPUs and accelerators, is taking over with its revolution in AI design.
When it comes to converting neural network programs into hardware, unlike other compilers, Apache TVM examines the “graph” of computer operations in a neural network and calculates how best to bring these operations to hardware based on dependencies between operations.
ApacheTVM is offered by OctoML, a company that helps operationalize AI. With Apache TVM, the company is therefore able to help companies optimize their neural networks to accommodate different hardware variants.
In his interview with ZDNet, OctoML co-founder and CEO Luis Ceze talked about how the technology is helping shape the development of artificial intelligence. According to Ceze, “Already in research, people are running model candidates through our platform, looking at the performance. The detailed performance metrics mean that ML developers can actually evaluate the models and pick the one that has the desired properties.”