quantization-aware-healing

Quantization-Aware Healing is a method that recovers compressed 4-bit language models by distilling directly from the original uncompressed model, offering faster and more stable performance than Quantization-Aware Training.