Can someone who has worked on this project write a better readme with commands and instructions to run the training and inference?