developer.nvidia.com
2026-06-27
NVIDIA has released the NVFP4 checkpoint for its Nemotron 3 Ultra model, leveraging the NVIDIA Model Optimizer to achieve efficient quantization. This advancement enables up to a 5.9x increase in inference throughput while maintaining BF16-level accuracy, significantly reducing hardware footprint. N
developer.nvidia.com
2026-06-09
NVIDIA introduces NVFP4 (NVIDIA Float Point 4) technology for large language model (LLM) pretraining, aiming to significantly enhance training efficiency while maintaining model accuracy. Utilizing sub-byte precision in JAX and MaxText frameworks, NVFP4 enables 4-bit mixed-precision pretraining on N