FisherSFT: Data-Efficient Supervised Fine-Tuning of Language Models Using Information Gain | Read Paper on Bytez