[ACL 2025] Outlier-Safe Pre-Training for Robust 4-Bit Quantization of Large Language Models - View it on GitHub