Summary: 17 instances, 15 unique

Text	Count
# TODO: Default args are not overwritten if provided as args	1
# FIXME: Wikilarge validations set only has 992 sentences	1
# TODO: Do we really need to extract the tokens	1
# TODO: Move this to setup or add the folders to the git repo	1
# FIXME: if the kwargs are already present in the args string, they will appear twice but fairseq will take only the last one into account	2
# TODO: Ideally we should lock the file when reading as well but it's slow	1
# TODO: Quick hack to remove empty directories	1
# TODO: Should be in ts.uts.training	2
log_ranks = [np.log(1 + len(get_word2rank()))]  # TODO: This is completely arbitrary	1
# TODO: do we really need to create an additional file for locking ?	1
# TODO: Locking a directory should lock all files in that directory	1
# TODO: Jointly mix files	1
# TODO: This is a redundant hack with what happens in fairseq_evaluate_and_save (predict_files and evaluate_kwargs), it should be fixed	1
query_embeddings = query_index.reconstruct_n(start_idx, actual_batch_size)  # TODO: Do this in the background	1
# TODO: Nested with lock_directory() should not be blocking	1