lower result for multi spans cases

[unbreading]

Hi minghao,

I try to use the pre-trained model to reproduce the result. But I got a lower result.

As shown above, all things go well but model performs really poor on multi-spans cases. I checked the output file (predictions.json), and I did not find a multi-spans prediction.

What I have done:

1. run run_mtmsn to evaluate with almost the same config as you mentioned. I only skipped the training process.
2. got an output file predictions.json
3. use drop_eval.py to evalute on predictions.json and drop_dataset_dev.json

Did I make any mistake? Please help me.
Thanks in advance.

[huminghao16]

Could you please show the running command?