One of the most important annual events in the field of artificial intelligence starts this week in Singapore: the International conference on learning representations. As usual, the giant of the Nvidia chip had a major presence at the conference, presenting more than 70 research articles from his team.
Also: Nvidia launches Nemo software tools to help companies create personalized AI agents
The articles cover subjects ranging from the generation of music to the creation of 3D realistic videos, robot training tasks and the ability to generate several important language models by pressing a button.
Not just a flea company
“People often consider Nvidia as a flea company that makes impressive fleas, and of course, we are really proud of it,” said Bryan Catanzaro, head of research applied in Deep Learning in Nvidia, in an interview with ZDNET. “But the history which, in my opinion, counts the most is that so that we can do these impressive chips, we must do research like this, because it teaches us to do all these systems.”
Also: Nvidia dominates in the Gen Ai references, the tokens of 2 chips Ai Rivaux
The articles presented this week, most of which have been published in the past year about the Arxiv preparation server, range from pure research to programs that offer immediately usable tools.
In the first category, for example, A project called llamaflex Improves the task of generating many models of important languages from a single parent. It is common today to “distill” only one, very large LLM in LLM “Student” which inherit the capacity of the “teacher” but take less memory storage.
Nvidia Ruisi Cai researchers and her team observed that the distillation method could be improved using what they call “elastic pre-training”. Taking a single large pre -formulated LLM – In this case, Llama 3.18b of Meta Platforms – they put it through a single additional training phase with 60 billion new training tokens. The result is an algorithm called “router” which can automatically generate any number of LLM of progeniture of different size to practically press a button.
FUGATTO Foundation model
In the category of more tangible programs, The Fugatto 1 program is a “foundation model” for audio synthesis, an AI model which can manage any combination of text instructions and sound clips and transform the instructions based on instructions. “I am really excited by Fugatto,” Catanzaro told Zdnet.
Also: I tried a lot of AI images generators, and Nvidia and Mit’s is the one to beat for the speed
For example, Fugatto can produce a sound on request, such as the meowing of a cat. He can choose a song sample to reproduce each separate singer. It can merge the corrugating water sound with the sound of a classic guitar to create a hybrid sound which is a mixture of the two.
Fugattto’s neuronal net is A developed in Google in 2022 This can work on “spectrograms”, sounds like wave models. The original contribution of Rafael Valle de Nvidia and his team is a new set of data and a training regime that teaches the model to manage complex textual orders.
Also: The rise of PCs: how companies reshape their technology to continue
Nvidia projects such as Fugatto are based on many previous innovations, just like any research laboratory. One of the important aspects that distinguishes NVIDIA’s research articles is that they tend to offer more technical details on the material implementations used in research, such as, for example, the number of GPU chips used, while other laboratories often leave this data.
IA informing the development of fleas
Research projects like Llamaflex and Fugatto perform many functions. They highlight the many ways in which Nvidia chips can be used, which is always a great way to promote the capacities of these parts. They also maintain Nvidia involved in the state of the art for AI, which can shed light on the development of the flea of the company. They help the company attract talent by presenting projects that can earn awards and peer recognition.
Also: Google’s latest chip is to reduce a huge cost hidden in AI
And they show how the raw power of Nvidia chips plays a big role in AI as a field. The “acceleration” of AI is a story that is not known as much as it should be, said Catanzaro.
“I believe that a lot of progress in AI in the past 15 years come from acceleration,” said Catanzaro.
Discover all NVIDIA research publications on The main research site.
Get the best morning stories in your reception box every day with our Newsletter Tech Today.