Join C-suite executives in San Francisco July 11-12 to hear how leaders are integrating and optimizing their AI investments for success.. learn more
The big artificial intelligence (AI) news at Google I/O today was the announcement of the company’s PaLM 2 large-scale language model, but that wasn’t the only AI news at the event.
The company is also rolling out a series of open source machine learning (ML) technology updates and enhancements for the growing TensorFlow ecosystem. TensorFlow is an open-source technology initiative led by Google that provides ML tools that help developers build and train models.
Google announces new DTensor technology at Google I/O. This technology brings new parallelism techniques to ML training and helps improve model training and scaling efficiency.
There is also a preview release of the TF Quantization API. This is intended to increase resource efficiency and reduce development costs across the model.
event
transform 2023
Join us July 11-12 in San Francisco. There, he shares how management integrated and optimized his AI investments to drive success and avoid common pitfalls.
Register now
A key part of the TensorFlow ecosystem is the Keras API suite, which provides a set of Python language-based deep learning capabilities in addition to the core TensorFlow technology. Google announced his two new Keras tools, KerasCV for computer vision (CV) applications and KerasNLP for natural language processing (NLP).
Alex Spinelli, vice president of machine learning product management at Google, said, “Most of what we see in the tools and open source space is really driving new capabilities, new efficiencies, and new performance. ‘” he told VentureBeat. . “Undoubtedly, Google will incorporate great AI and ML into their products, but we also want to create a rising tide that lifts all boats, so we are serious about our open source strategy. We are committed to helping developers across the board.”
TensorFlow remains the machine learning ‘workshop’ at Google
Spinelli emphasized that in an era where large language models (LLMs) are all the rage, having the right ML training tools is more important than ever.
“TensorFlow is still a workhorse in machine learning,” he said. “It’s still … basic underlying infrastructure.” [in Google] It drives much of our own machine learning development. ”
To that end, DTensor updates provide more “horsepower” as ML training requirements continue to grow. DTensor introduces more parallelization features to help optimize your training workflow.
Spinelli said ML as a whole is driving an even greater hunger for data and computing resources. Therefore, it is very important to find ways to improve performance to handle more data to meet the needs of increasingly large models. The new Keras update offers even more power with modular components that allow developers to actually build their own computer his vision and natural language processing capabilities.
TensorFlow gets even more powerful thanks to the new JAX2TF technology. JAX is an AI research framework and is widely used as a computational library at Google to build technologies such as the Bard AI chatbot. JAX2TF makes it easier for models written in JAX to be used in the TensorFlow ecosystem.
“One of the things we’re really looking forward to is seeing how these things get into the product and see that developer community thrive,” he said.
PyTorch and TensorFlow
TensorFlow is the flagship product of Google’s ML efforts, but it’s not the only open-source ML training library.
In recent years, the open-source PyTorch framework, originally created by Facebook (now Meta), has grown in popularity. In 2022, Meta donated his PyTorch to the Linux Foundation to create his new PyTorch Foundation, a multi-stakeholder initiative with an open governance model.
Spinelli said what Google is trying to do is support developer choice when it comes to ML tools. He also said that TensorFlow is not just an ML framework, but a whole ecosystem of tools for ML that help support training and development for a wide range of use cases and deployment scenarios.
“It’s essentially the same set of technologies that Google uses to build machine learning,” Spinelli said. “If you really want to build large, high-performance systems and want to know that they will work in all future infrastructures, I think we have a very competitive offering. increase.”
One thing Google clearly won’t do is follow Meta’s lead and create an independent TensorFlor Foundation organization.
“We’re pretty happy with how we’re developing and managing today,” Spinelli said. “We are very happy with some of the great updates we are releasing now.”
VentureBeat Mission will be the digital town square for technical decision makers to gain knowledge about transformative enterprise technologies and transactions. Watch the briefing.
