From NVIDIA: 2024-04-18 12:30:23

NVIDIA announces optimizations for Meta Llama 3 with AI accelerations. Meta trained Llama 3 on 24,576 NVIDIA H100 GPUs and plans to scale its infrastructure to 350,000 GPUs. Llama 3 is available for cloud, data center, edge, and PC use. Businesses can fine-tune Llama 3 with NVIDIA NeMo for custom models. Llama 3 also runs on NVIDIA Jetson Orin for robotics and edge devices, and NVIDIA RTX GPUs for workstations and PCs. NVIDIA offers optimal performance with Llama 3 chatbots using metrics like tokens generated per second for cost-efficiency. NVIDIA advances community models to promote AI transparency and resilience.



Read more at NVIDIA: NVIDIA Accelerates Inference on Meta Llama 3 | NVIDIA Blog