NVIDIA has open-sourced two new large models, Nemotron-4-Minitron-4B and Nemotron-4-Minitron-8B, which utilize structured pruning and knowledge distillation for efficient training. This significantly reduces training requirements, minimizing data and computing power consumption. Compared to traditional methods, the new models reduce training token data by 40 times and save 1.8 times in computing costs. By optimizing Llama-3.18B, structured pruning simplifies the model structure, while knowledge distillation enhances performance.

Three authors have sued Nvidia for using their copyrighted books without permission to train the NeMo AI platform. The lawsuit claims that the authors' works became part of a dataset consisting of approximately 196,640 books. The authors assert that their works were removed after training in NeMo. Nvidia has not issued a formal response to the case, which could set a significant precedent for copyright in the AI field.

The Nvidia NeMo project demonstrates the application of generative AI in semiconductor chip design, training large language models on internal data to enhance productivity. The research highlights the potential of Nvidia NeMo to provide customized AI models in the semiconductor sector, offering a competitive edge. Nvidia's custom large language model, ChipNeMo, has applications that include chatbots, code generators, and analytical tools, providing a comprehensive boost to semiconductor design.