Optimizing Exponent Bias for Sub-8bit Floating-Point Inference of Fine-tuned Transformers | IEEE Conference Publication | IEEE Xplore