FP4 All the Way: Fully Quantized Training of Large Language Models | Read Paper on Bytez