Fetching the paper…

Quantization Variation: A New Perspective on Training Transformers with Low-Bit Precision · Around