OctoML builds on the Apache TVM open-source machine learning compiler to automatically optimize models so they can run on virtually any hardware. Its platform delivers model performance improvements of 2–10x for customers and is used by Global 100 companies, including Toyota. The company has signed hardware partners including Qualcomm, AMD and Arm and has worked with Microsoft on large-scale video content moderation deployments. OctoML has been onboarding early adopters to a SaaS deployment platform that has not yet reached general availability but is expanding its customer base. The company says it has a healthy runway and plans to use new funding to scale engineering and sales teams and to build out its partner ecosystem. Co-founders Luis Ceze, Tianqi Chen, Jason Knight, Jared Roesch and Thierry Moreau together created the Apache TVM framework. Seattle-based OctoML provides a machine-learning acceleration platform built on the open-source Apache TVM compiler that automatically optimizes models for cloud and edge devices. Its flagship SaaS, Octomizer, is in early access with close to 1,000 waitlist sign-ups and several paying customers. The founding team created TVM and OctoML has partnerships with Microsoft (also a customer), Qualcomm, and AMD; it recently added support for Apple M1. Management plans to expand engineering, accelerate go-to-market, and build a customer success team while adding deployment support and end-to-end training optimization. These product efforts aim to speed up training, reduce costs, and simplify model deployments for users. Financially, OctoML has raised a total of $47M to date, including its new Series B and a prior $15M Series A. OctoML commercializes the Apache TVM machine learning compiler stack to automatically optimize and package ML models for target hardware. Its SaaS product, the Octomizer, lets users upload models to automatically optimize, benchmark and package them and can be integrated via API into CI/CD pipelines. The underlying TVM project is an Apache incubating project with usage and support from AWS, ARM, Facebook, Google, Intel, Microsoft, Nvidia, Xilinx and others; TVM already powers Amazon Alexa’s wake word detection and can yield up to 80x performance gains in some cases. The company was formed by the team behind TVM and is led by CEO and University of Washington professor Luis Ceze. OctoML currently employs about 20 engineers. With the new funding the company plans to expand its engineering team, hire an evangelist, and evolve the Octomizer into a fuller MLOps platform. OctoML offers a managed service that automates optimization, compilation and secure deployment of deep learning models across a broad set of hardware targets in cloud and edge environments. It leverages Apache TVM, an open-source model optimization and compilation stack originated by the founding team, which has more than 270 contributors and is used by companies such as Amazon, Facebook, Microsoft, Xilinx, and Qualcomm. The company aims to reduce engineering and operating costs and the risk of dependence on specific platforms by using machine-learning based code generation to automate per-hardware optimization. OctoML plans to ramp up work with hardware vendors to optimize the TVM stack for current and upcoming chips and to keep deployed models up and running at peak efficiency. The founding team includes Luis Ceze (CEO), Tianqi Chen (CTO), Jason Knight (CPO), Thierry Moreau and Jared Roesch, and the company counts Carlos Guestrin as an advisor. OctoML is a spinout of the University of Washington Allen School and is based in Seattle.

Total raised

$132M

Funding rounds

4

Latest round

Series C

Latest activity

Nov 2021