AngledLuffa
commented
4 years ago Are you comfortable using branches? I think this should be fixed in the fix_mwt branch. Thanks for reporting.
Closed sarves closed 3 years ago
AngledLuffa
commented
4 years ago Are you comfortable using branches? I think this should be fixed in the fix_mwt branch. Thanks for reporting.
sarves
commented
4 years ago @AngledLuffa Thank you for fixing. So I can just replace new stanza/stanza/models/mwt/trainer.py from dev branch and try . Will do and revert back.
Describe the bug I tried to train Stanza for Tamil, and mwt training always (tried with different data set) breaks at 33rd epoch. Log: 2020-08-20 09:24:11: step 360/1100 (epoch 33/100), loss = 0.286464 (0.054 sec/batch), lr: 0.001000 Evaluating on dev set... epoch 33: train_loss = 0.271402, dev_score = 0.9390 Traceback (most recent call last): File "/usr/lib/python3.8/runpy.py", line 193, in _run_module_as_main return _run_code(code, main_globals, None, File "/usr/lib/python3.8/runpy.py", line 86, in _run_code exec(code, run_globals) File "/home/sarves/Stanza/stanza-train-master/stanza/stanza/models/mwt_expander.py", line 255, in
main()
File "/home/sarves/Stanza/stanza-train-master/stanza/stanza/models/mwt_expander.py", line 89, in main
train(args)
File "/home/sarves/Stanza/stanza-train-master/stanza/stanza/models/mwt_expander.py", line 182, in train
trainer.change_lr(current_lr)
AttributeError: 'Trainer' object has no attribute 'change_lr'
Running MWT expander in predict mode
Building an attentional Seq2Seq model...
Using a Bi-LSTM encoder
Using soft attention for LSTM.
Finetune all embeddings.
max_dec_len: 39
Loading data with batch size 50...
3 batches created.
Running the seq2seq model...
/pytorch/aten/src/ATen/native/BinaryOps.cpp:81: UserWarning: Integer division of tensors using div or / is deprecated, and in a future release div will perform true division as in Python 3. Use true_divide or floor_divide (// in Python) instead.