rosmine

Standard LLM training is surprisingly bad at making the model outputs match the training data distribution, so the writing quality is bad. I made a new training algorithm called Distribution Fine Tuning (DFT) to fix…