3 ms·Nvidia: Quantization-Aware Distillation for NVFP4 Inference Accuracy Recovery [pdf]4 points by tosh 10mo ago