single mention · inference & serving
Quantization-aware distillation
Quantization-aware distillation trains a quantized student model, initialized from a post-training-quantized checkpoint, to recover accuracy lost to quantization.1 The student is distilled from a quantization-reference checkpoint, with the quantization configuration taken from that reference.2
How this definition was written
Summarised by glm-5.3-flash from 1 source; every sentence is checked against the quote it cites.
We initialize each QAD student from a post-training quantized (PTQ) checkpoint
— NVIDIA Nemotron 3.5 Super and Lightning Technical ReportWe derive the quantization configuration from an existingquantization- reference checkpoint
— NVIDIA Nemotron 3.5 Super and Lightning Technical Report
Also called Quantization-aware distillation for NVFP4.
Filed under inference & serving › inference quantization