mcao516 / entfa Goto Github PK

In the paper you say:

For training the CMLM, we use both XSUM and the CNN/Dailymail dataset.

Do you train the CMLM separately for each dataset depending on the downstream task, or do you train one CMLM using both datasets?

Was the conditional masked language model checkpoint in the description used to generate the summaries included in data/test.json / data/train.json? If not, would it be possible to have the checkpoint that you used for generating summaries?

Hi, will you provide the instruction for experiment reproduction and code structure explanation? Thanks.

Hi, I found the follwing data samples in xsum test set will meet assert position[1] in end_pos, "- {}\n- {}\n- {}\n- {}\n- {}\n".format(position, tokens, probs, entity, end_pos) error
(

[240, 349, 398, 646, 679, 957, 1003, 1293, 1501, 1516, 2006, 2069, 2123, 2811, 3160, 3354, 3679, 4015, 4102, 4833, 5392, 5852, 6214, 6564, 6915, 6930, 7379, 7481, 8043, 8159, 8718, 9473, 9648, 9763, 10456, 10717]

Hi again! Would you be able to share the summaries that were generated by the best model reported in your paper? We are reproducing and comparing our own approaches with your results, so it would be great to have the summaries to verify that we're on track :)

Thanks,
Anton