OctoML

About OctoML

Based out of Seattle, Washington, OctoML was founded by the team that created Apache TVM, an open source machine learning compiler that optimizes models so they can run efficiently in the cloud or on edge devices (CPUs, GPUs, NPUs, Accelerators), saving time, energy and cost.

Our customers upload their trained deep learning models (from TensorFlow, PyTorch, Keras, ONNX, MxNet, etc) to the Octomizer (our SaaS product based on Apache TVM), which outputs an optimized and packaged version of the model fine-tuned to the target hardware. Our platform also benchmarks the model on various target hardware options, allowing you to choose the most cost efficient device (while meeting latency and throughput requirements) for your use case.