Quantization-aware training (QAT), low-bit LLMs (INT4, INT3, INT2, NVFP4), quantization-aware distillation
QUASAR: Lowering the Loss Floor of Quantization-Aware Training with Loss-Aware Reconstruction