Quantize transformers to any learned arbitrary 4-bit numeric format - View it on GitHub