👋 Need help with code?
QUASAR: How Saliency-Weighted Reconstruction Closes the Loss Floor Gap in LLM Quantization-Aware Training