jiachenwestlake / entity_bert Goto Github PK

作者您好。百忙之中打扰您，我想请问一下咱们的数据是怎么进行预处理的，方便分享一下源码吗？

hi, thanks for your job!
I download your pre-trained BERT weights (news) and fine-tune on NER task. BUT i got f1=r=p=0
Do you have any idea?

hi, thanks for your job!, when I was doing pre-training task (download BERT-base weights & THUCNews dataset), I set the batch to 1, and I found that the value of loss fluctuated after 10,000 steps (test by cpu) and did not decrease all the time, Is this normal?

What GPU do you use and how long did it take to complete the pre-training? thank you！

log:
12/18/2020 03:58:32 - INFO - main - Num of trianing step: 4092, steps loss: 4.7691, steps entity right rate: 0.0312
12/18/2020 03:58:37 - INFO - main - Num of trianing step: 4093, steps loss: 5.0219, steps entity right rate: 0.0476
12/18/2020 03:58:41 - INFO - main - Num of trianing step: 4094, steps loss: 3.6677, steps entity right rate: 0.0968
12/18/2020 03:58:46 - INFO - main - Num of trianing step: 4095, steps loss: 4.1704, steps entity right rate: 0.0789

12/18/2020 15:07:20 - INFO - main - Num of trianing step: 12157, steps loss: 5.3570, steps entity right rate: 0.1818
12/18/2020 15:07:25 - INFO - main - Num of trianing step: 12158, steps loss: 4.6864, steps entity right rate: 0.0909
12/18/2020 15:07:30 - INFO - main - Num of trianing step: 12159, steps loss: 3.1328, steps entity right rate: 0.1818
12/18/2020 15:07:36 - INFO - main - Num of trianing step: 12160, steps loss: 4.6794, steps entity right rate: 0.1273
12/18/2020 15:07:41 - INFO - main - Num of trianing step: 12161, steps loss: 3.4541, steps entity right rate: 0.1739
12/18/2020 15:07:46 - INFO - main - Num of trianing step: 12162, steps loss: 2.7492, steps entity right rate: 0.2333
12/18/2020 15:07:52 - INFO - main - Num of trianing step: 12163, steps loss: 4.6251, steps entity right rate: 0.2830

您好，在运行train.py的时候提示AttributeError: 'NoneType' object has no attribute 'batch_step'，想请问一下怎么修正。

01/24/2021 11:43:41 - INFO - utils.tokenization - loading vocabulary file https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-uncased-vocab.txt from cache at C:\Users\z1486.pytorch_pretrained_bert\26bc1ad6c0ac742e9b52263248f6d0f00068293b33709fae12320c0e35ccfbbb.542ce4285a40d23a559526243235df47c5f75c197f04f37d1a0c124c32c9a084
Traceback (most recent call last):
File "F:/LectureArtical/Entity-BERT/Entity_BERT-master/train.py", line 143, in
total_train_steps = int(train_iter.batch_step * args.num_train_epochs)
AttributeError: 'NoneType' object has no attribute 'batch_step'

作者您好，感谢您的无私分享和贡献，我想请问一下，在此finance dataset下 ../data/dataset_finance/ner/annual_report_anno_part1&2_title.txt怎么数据集下载文件当中没有呢？可以给解答一下吗？叨扰大佬了