NVIDIA explains NVFP4 for cheaper LLM inference

NVFP4 is a 4-bit floating-point format that reduces memory use and speeds up LLM inference. Videos demonstrate creating NVFP4-quantized Nemotron 3 Ultra checkpoints using NVIDIA Model Optimizer.