I am training a coreference resolution model with Alennlp. However after I execute the command train_data_path=<train path> validation_data_path=<dev path> test_data_path=<testpath> dataset_reader=<dataset reader> allennlp train <jsonnet file> -s <output dir> --include-package <training module>
it does not produce a model.tar.gz file in the <output dir>.
Any idea why?
edit: I've been running this via polyaxon and somehow the error logs does not appear after it fails. Rightfully once training completes the model.tar.gz file should appear. I think it should be an error that appears during training that fails it but I am unable to inspect it
The
Trainsubcommand will always produce themodel.tar.gzarchive so long as (1)--dry-runis not supplied to the CLI; and (2) the training process ran to completion without error (cf. code). You likely are having an error during training that needs to be corrected.