Neuropod: An Open-Source, Framework-Agnostic Inference Engine for Taking Research to Production
, Uber Advanced Technologies Group
Learn how to quickly productionize deep learning models from any framework. Although it may be straightforward for researchers to experiment with several DL frameworks, adding production support for a new framework is no small task. To solve this, we built Neuropod, an abstraction layer on top of deep learning frameworks that provides a uniform interface to run DL models. It enables seamless model deployment that's agnostic of the framework in which the model was implemented. Neuropod was created at Uber ATG to help support our self-driving efforts and has been instrumental in quickly deploying new models. We've deployed hundreds of Neuropod models across ATG, AI Labs, and the Uber core business. We’ll go over how Neuropod works, why we built it, and how it makes it easier to productionize research. We’ll also talk about some new capabilities that this open-source library enables.