mingzhu0527 / mashqa Goto Github PK

Where can I find a pre-trained model, and how do I use it for inference?

during running preprocessing script on google colab
!sh scripts/prepro_sentqa.sh

I got below error
Traceback (most recent call last):
File "run_sentqa_att.py", line 1050, in
tf.app.run()
AttributeError: module 'tensorflow' has no attribute 'app'

I tried import tensorflow.compat.v1 as tf but error still exists

Can you tell me how to fix this issue and what is tensorflow version you used ?

Thanks

Could you tell what the first two columns signify in the prediction file while the other two being predicted_label and ground_truth_label respectively (total 4 columns are present) ?

When training the model using the default settings the training loss goes to zero within 100 training steps. I'm using tensorflow-gpu 1.14. I've verified that the input data is read correctly.

Can you tell me how to fix this issue?

Hi!

I am confused about the ['sent_starts'] in the data files:

data["data"][3]["paragraphs"][0]["sent_starts"]
[[0, 46], [47, 76], [124, 110], [235, 484], [720, 679], [1400, 85], [1486, 24], [1511, 53], [1565, 226], [1792, 30], [1823, 123], [1947, 20], [1968, 192], [2161, 77], [2239, 152], [2392, 147], [2540, 46], [2587, 51], [2639, 63], [2703, 121], [2825, 265], [3091, 218], [3310, 105], [3416, 279]]

I would have assumed that the first element of each sublist is the start of the sentence in 'context', and the second element would be the end. Therefore, the second element should be a higher number than the first element. However, starting from [1400, 85], this is not the case anymore. I really wonder why this is the case.

Thanks a lot!

Does this model have multi-lingual support? I want to infer it on a Bengali multispan dataset.

And I also want to know whether it will be a good idea to fine tune this model for bengali- multispan question answering on another specific domain.

Sorry if the question is too silly as I'm still a novice at NLP.

Hi there, I had a doubt regarding the val and test sets for this dataset.

The paper in Table 1 reports that there are 3,493 QA pairs in the validation split and 3,587 QA pairs in the test split. However, upon examining the provided documents in the google drive, I found that there are actually 3587 QA pairs in the validation split and 3493 QA pairs in the test split. I observed the same discrepancy for the consec version of the files as well. Could you clarify if this might be a naming issue for the files?

Thanks!

Sorry to have this naive question. I have already download the code here but found no dataset attached with it. Where can I download the entire MASH-QA dataset? Did I miss it somewhere?

When doing sh sentqa_GPU.sh , I get this error OSError: Not found: "xlnet_cased_L-24_H-1024_A-16/spiece.model": No such file or directory Error #2`.
Can you tell how can I solve this?