Continual Quantization-Aware Pre-Training: When to transition from 16-bit to 1.58-bit pre-training for BitNet language models? Paper • 2502.11895 • Published Feb 17 • 3
