OctoML

Automate your deep learning engineering.

Ready for state-of-the-art model performance, powered by Apache TVM, at your fingertips? Get in touch for Octomizer Early Access.

Why OctoML?

OctoML applies cutting edge AI to make it easier and faster to put machine learning models into production on any hardware.

From the team that built the Apache TVM deep learning compiler stack, used in production by top ML and hardware vendors, OctoML is automating the deployment of machine learning models.

Forget what you think you know about ML deployment.

OctoML provides an automated, streamlined process.

Easy

Optimize and deploy higher performing, hardware-portable ML models through our web UI or API - it's that simple.

Fast

Reduce model deployment time from months to hours and get your ML model to market in a fraction of the time.

Portable

We’re all about flexibility. Implement your machine learning models on any hardware, in the cloud, or on the edge.

HOW IT WORKS

Machine learning for your machine learning

Here's how we Octomize your model.

Optimize

We specialize the output on your particular topology for each hardware platform separately to ensure last mile optimization and efficiency.

Benchmark

Wonder how your now-optimized model stacks up against other common models? In other clouds? On the edge? We’ll show it all.

Package

Your trained ML models will be ready for simple, secure, and efficient deployment in your edge and cloud environments.

Try before you buy

Looking to buy hardware for your ML project? Use our comprehensive benchmark to buy smart.

How we do it

Meet the Octomizer.

Powerful software for easier ML deployments.

The Octomizer is a software as a service (SaaS) product designed to make it easy for anyone to deploy ML models as part of their applications. Including the optimization, benchmarking, and packaging steps critical to any production deployment.

Also, use Octomizer as a sizing tool to help determine your next machine learning hardware purchase decision based on easily accessible and comparable perf/watt and perf/$ metrics

Step 1

Upload a model topology in whatever format you have (TensorFlow, PyTorch, ONNX, etc.).

Step 2

We optimize, benchmark, and package the model across a number of hardware platforms and application language runtimes.

Step 3

Easily compare performance of a model across various cloud CPU and GPU instance types and evaluate the device sizing requirements needed to deploy your models on ARM mobile or embedded processors.

Step 4

Choose from a variety of deployment packaging formats such as Python wheel, shared library with C API, serverless cloud tarball, and others.

Step 5

Receive your binary and deploy your model using whatever deployment machinery you currently use.

We’re automating AI and ML ops through a unified software foundation.

Machine learning software infrastructure can be complex and clunky. We're making it easy, efficient, and ensuring it covers the hardware that you are about.

Compilers over kernel libraries

With a heavy reliance on handwritten and optimized kernel libraries, machine learning software systems are often inflexible in operator and hardware coverage. So four years ago, the Apache TVM research project was created out of the University of Washington to make efficient and portable machine learning possible through a compiler approach. This project is now used in production at Microsoft, Amazon, Facebook, and many more companies.

Systems over SoTA

Much of the machine learning press focuses on new models and techniques achieving “state of the art” (SoTA) results or performance, which has led machine learning systems to be stuck in a rut. At OctoML, we go beyond achieving one-off SoTA results to deliver robust, intelligent systems that embrace automation and are more applicable to your problem, in your domain, on your hardware.

We're on a mission to overhaul the way ML models are optimized and deployed by bringing the latest in research and open source to more users and more hardware platforms.

Meet our team of Octonauts

About our World-class team

The current Cambrian explosion of machine learning algorithms, use cases, and hardware has begun to strain existing ML software stacks to the limit.

With decades of combined experience in computer systems design and machine learning, we believe automated systems are the right approach to reign in the complexity and enable us all to move forward more easily.

Our team is composed of passionate ML PhDs, pioneers and professors with experience at Microsoft, Facebook, Amazon, Apple, Qualcomm, Intel and more.

From the creators of