Annotation order in training.jsonl

Is the order of data in training.jsonl file in the order they were annotated in the UI? Is there a way we can add a timestamp to the dataset?

The data that’s exported to training.jsonl after training is the exact data the model was trained on. This means that the data was shuffled, split into training / evaluation examples, split into sentences (if auto sentence segmentation was enabled) and with merged "spans" (if you’re using ner.batch-train). So it can be potentially be quite different from the data in the original datasets and its order will be completely random, because it was shuffled multiple times in the training loop.

If you want a record of the annotated examples in order, you probably want to export the dataset using the db-out command. This should give you the dataset in order (ascending for SQLite, descending for Postgres – see here for details). The examples themselves do not have timestamps (although you could probably hack in a batch timestamp by modifying the /give_answers endpoint). But each annotation session will be saved as a separate, timestamped dataset.