The move comes as artificial intelligence companies such as OpenAI, Google and Anthropic develop their own increasingly proficient chips, and could act as a hedge against those companies moving away from Nvidia’s technology over time.
Open models are already a key part of the AI ecosystem, with many researchers and startups using them to experiment, prototype and build. While OpenAI and Google offer diminutive open models, they don’t update them as frequently as their rivals in China. For this and other reasons, open models of Chinese companies are now much more popular, he says data from Hugging Facehosting platform for open source projects.
According to benchmark results released by the company ahead of launch, the novel Nvidia Nemotron 3 models are among the best you can download, modify and run on your own hardware.
“Open innovation is the foundation of AI progress,” CEO Jensen Huang said in a statement ahead of the news. “With Nemotron, we are transforming advanced AI into an open platform that gives developers the transparency and efficiency they need to build agent systems at scale.”
Nvidia is taking a more crystal clear approach than many of its US rivals by sharing the data it uses to train Nemotron – a fact that should assist engineers tweak models more easily. The company also provides tools to assist with customization and tuning. This includes a novel hybrid hidden model architecture consisting of a mix of experts, which Nvidia says is particularly good at creating AI agents that can take actions on computers or on the Internet. The company is also launching libraries that allow users to train agents to perform actions using reinforcement learning, which involves providing models with simulated rewards and punishments.
Nemotron 3 models come in three sizes: Nano, which has 30 billion parameters; Super, which is 100 billion; and Ultra, which has 500 billion. The model’s parameters loosely correspond to its capabilities and awkwardness to utilize. The largest models are so bulky that they must run on racks of steep equipment.
Model basics
Kari Ann Briski, vice president of generative AI enterprise software at Nvidia, said open models are essential to AI developers for three reasons: Designers increasingly need to adapt models to specific tasks; forwarding queries to different models often helps; furthermore, after training, it is easier to squeeze more wise responses out of these models by asking them to perform some type of simulated reasoning. “We believe that open source is the foundation of AI innovation, continually accelerating the global economy,” Briski said.
Social media giant Meta released its first advanced open models under the name Llama in February 2023. However, as competition intensifies, Meta has signaled that its future releases may not be open source.
This move is part of a broader trend in the AI industry. Over the past year, American companies have become less open, more secretive about their research and reluctant to tell their rivals about the latest engineering tricks.