Earlier this week, Nvidia released a new artificial intelligence model, Nemotron, with benchmark tests showing the model outperforming offerings from industry heavyweights, OpenAI and Anthropic.
The product’s full name is Llama-3.1-Nemotron-70B-Instruct and it has similarities to Meta’s Llama 3.1 70B offering. It debuted on the wide-ranging AI hosting platform HuggingFace – which recently surpassed one million models – without much fuss but its performance has ensured word is spreading fast.
The open-source reward model has its training dataset available on HuggingFace, with a test preview on the company website.
Nvidia has reported its new release has achieved top scores across various markers, including 85.0 on the Arena Hard benchmark, 57.6 on AlpacaEval 2 LC, and an 8.98 score on the GPT-4-Turbo MT-Bench.
Those results compare very favorably to established models such as OpenAI’s ChatGPT-4o as well as Anthropic’s Claude 3.5 Sonnet offering, even though Nemotron is considerably smaller at 70B parameters.
Our Llama-3.1-Nemotron-70B-Instruct model is a leading model on the 🏆 Arena Hard benchmark (85) from @lmarena_ai.
Arena Hard uses a data pipeline to build high-quality benchmarks from live data in Chatbot Arena, and is known for its predictive ability of Chatbot Arena Elo… pic.twitter.com/HczLQQ6EOp
— NVIDIA AI Developer (@NVIDIAAIDev) October 15, 2024
Diversifying beyond its core GPU market
Not only do the performance markers indicate high scores, they also catapult Nvidia’s reputation and growing influence in AI development, beyond its core focus on GPUs.
The American multinational, headquartered in Santa Clara, California, is showing that it can diversify beyond its dominant market to make a significant impact in AI software. It has also struck a blow for the fledgling AI developers against the big boys in the sector.
Nvidia has also recently agreed terms with Taiwan’s Foxconn to build the world’s largest superchip production facility in Mexico.
As stated above, Meta’s Llama 3.1 was a key influence on Nvidia’s project with the company refining the existing product. Using advanced techniques including Reinforcement Learning from Human Feedback (RLHF), Nvidia sought to train its AI model to provide a greater level of performance, offering a cost-effective resource with stronger capabilities than some of the main models available on the market.
Image credit: Via Ideogram