Walk me through quantization techniques for inference, including INT8, FP8, and INT4, for NVIDIA deployments.
