Victoria X Lin
Victoria X Lin
I have met exactly the same issue. Were you able to work around it?
This is a simple workflow that trains a sequence-to-sequence model. And this is how I'm using the function. ```# Monitor training with Tensorboard writer = SummaryWriter(log_dir=get_log_dir('tensorboard', self.args)) for epoch_id in...
No, there wasn't. Should I simply use `add_graph` after the writer was created?
Echo the message above. The issue remained unsolved for me.
Varying the batch size your way shouldn't change performance because the effective batch size is kept at 32. Would you mind sharing your training curve or the accuracy change log?
The difference might be caused by the **Data Repair** step described in section 4.3 of the paper. I've modified the data processing steps to incorporate this step, please `git pull`...
Sorry I introduced this bug with the checkpoints release. A temporary fix for Spider is to comment out line https://github.com/salesforce/TabularSemanticParsing/blob/main/src/utils/trans/bert_utils.py#L31 and uncomment line https://github.com/salesforce/TabularSemanticParsing/blob/main/src/utils/trans/bert_utils.py#L30. The issues is that in the...
I will push a more stable fix later.
Try replacing the BERT model we used with a multilingual LM such as mBERT or XLM-R. They can be accessed the same way via Hugging Face transformers library.
You mean changing action dropout rate from 0.0 to 0.1? 0.9 is very aggressive dropout and 1.0 implies dropout everything and randomly sample an edge. If so 0.1 dropout rate...