checkpoint-1600/training_state.json

{"global_step": 1600, "epoch_step": 1866, "epoch": 267, "exhausted_backends": [], "repeats": {"training-images": 0}}