Google releases tools for AI model development

2 mins

Greg Russell

Published by: Greg Russell

10 April 2024, 01:07PM GMT+00:00

In Brief

Google introduces open-source tools for AI model development.

JetStream promises 3x higher performance per dollar.

Collaboration with Hugging Face yields Optimum TPU.

Tools aim to optimize AI workloads on TPUs and Nvidia GPUs.

Initiative represents Google's commitment to AI innovation.

Google releases tools for AI model development

Google has introduced several key open-source applications designed to bolster generative AI projects and infrastructure at its developer conference, Cloud Next. The first is MaxDiffusion, a set of diffusion model reference implementations that have been optimized for XLA devices including Google’s tensor-processing units and Nvidia’s most recent GPUs. The next release, JetStream, will deliver up to three times more performance per dollar for text-generating models such as Gemma 7B and Llama 2, and will initially support TPUs while adding GPU support later on.

JetStream is essentially an engine at the heart of Google’s offerings. Resketched to work with TPUs now but planned to cooperate with GPUs sooner, JetStream’s objective is to be a cost-effective inference stack for rendering AI workloads in real conditions. Mark Lohmeyer, the general manager of Google Cloud’s compute and machine learning infrastructure, said that the system will offer the potential for a high level of performance and cost-effectiveness. However, the focus will still rely on the most popular open models, such as Gemma 7B, and Llama 2. Google’s collaboration with Hugging Face birthed Optimum TPU, a tool that makes building workspaces with generative AI models feasible on TPUs. For now, Optimum TPU would not assist all models in the Hugging Face operation.

Compared to previous experiences, Google’s prospects in open-source AI tools are a dramatic departure from before, and the intention is to win the developers’ hearts and become a complete ecosystem. In this, MaxDiffusion, JetStream, and Optimum TPU demonstrate that Google is eager to support AI model creation and the infrastructure on which models run. Three Keynotes that emphasize cost-effectiveness and efficiency are most likely shown to be central in allowing developers to engage generative AI innovations for their work when the AI landscape is transformed.

Google's foray into open-source AI tools underscores its dedication to fostering innovation and collaboration within the AI community. By democratizing access to advanced AI capabilities, Google aims to accelerate the development and deployment of cutting-edge AI solutions across industries.



Blue robot
Brown robot
Green robot
Purple robot

Share this material in socials

Copy link
Bell notification
Blue mail
Blured bell
Blue Mail
Mail plane
Mail plane
Mail icon
Mail icon
Mail icon

Join our newsletter

Stay in the know on the latest alpha, news and product updates.