Build More Accurate and Efficient AI Agents with the New NVIDIA Llama Nemotron Super v1.5

Introduction

The NVIDIA Llama Nemotron Super v1.5 is introduced as an enhanced version of the NVIDIA Nemotron family, focusing on improving accuracy, efficiency, and transparency of AI agents. It is built on strong open models in the ecosystem and utilizes NVIDIA synthetic datasets, advanced techniques, and tools.

Core Enhancements

• The Llama Nemotron Super v1.5 excels in core reasoning and agentic tasks including math, science, coding, function calling, instruction following, and chat.
• It surpasses other models in its weight class, particularly in multi-step reasoning and structured tool use tasks.
• Post-training on a new high-signal reasoning dataset enhances the model's accuracy and efficiency.

Throughput and Efficiency

• The model has been optimized for higher throughput and deployment efficiency to explore complex problem spaces within the same compute and time budget.
• Pruning techniques like neural architecture search have been applied to reduce inference costs and improve reasoning speed.
• Llama Nemotron Super v1.5 is designed to run on a single GPU, further enhancing compute efficiency.

Try Now

Experience the capabilities of Llama Nemotron Super v1.5 by visiting build.nvidia.com or downloading the model directly from Hugging Face.