Continual Quantization-Aware Pre-Training: When to transition from 16-bit to 1.58-bit pre-training for BitNet language models?

Jacob Nielsen, Peter Schneider-Kamp, Lukas Galke

Browse the full ACL paper archive.