training a udify model only for Korean

[euhkim]

Hello, I am training a udify model only for Korean where I train only on the Korean data from UD2.3. However, I am running into the following issue. The same code runs fine on other languages from UD.
Traceback (most recent call last):
File "train.py", line 113, in
train_model(train_params, serialization_dir, recover=bool(args.resume))
File "/home/user/anaconda3/envs/dependency_parse/lib/python3.7/site-packages/allennlp/commands/train.py", line 226, in train_model
cache_prefix)
File "/home/user/anaconda3/envs/dependency_parse/lib/python3.7/site-packages/allennlp/training/trainer_pieces.py", line 65, in from_params
model = Model.from_params(vocab=vocab, params=params.pop('model'))
File "/home/user/anaconda3/envs/dependency_parse/lib/python3.7/site-packages/allennlp/common/from_params.py", line 365, in from_params
return subclass.from_params(params=params, **extras)
File "/home/user/anaconda3/envs/dependency_parse/lib/python3.7/site-packages/allennlp/common/from_params.py", line 386, in from_params
kwargs = create_kwargs(cls, params, **extras)
File "/home/user/anaconda3/envs/dependency_parse/lib/python3.7/site-packages/allennlp/common/from_params.py", line 133, in create_kwargs
kwargs[name] = construct_arg(cls, name, annotation, param.default, params, **extras)
File "/home/user/anaconda3/envs/dependency_parse/lib/python3.7/site-packages/allennlp/common/from_params.py", line 257, in construct_arg
value_dict[key] = value_cls.from_params(params=value_params, **subextras)
File "/home/user/anaconda3/envs/dependency_parse/lib/python3.7/site-packages/allennlp/common/from_params.py", line 365, in from_params
return subclass.from_params(params=params, **extras)
File "/home/user/anaconda3/envs/dependency_parse/lib/python3.7/site-packages/allennlp/common/from_params.py", line 388, in from_params
return cls(**kwargs) # type: ignore
File "/home/user/udify-master/udify/models/tag_decoder.py", line 105, in init
div_value=4.0)
File "/home/user/anaconda3/envs/dependency_parse/lib/python3.7/site-packages/torch/nn/modules/adaptive.py", line 133, in init
raise ValueError("cutoffs should be a sequence of unique, positive "
ValueError: cutoffs should be a sequence of unique, positive integers sorted in an increasing order, where each value is between 1 and n_classes-1

[dfvalio]

Hi,
I am facing the exact same problem with the following languages: Hindi, Thai, Chinese and Japanese. Have you been able to solve your issue? If so, could you please share the solution?
Thank you in advance.

[ziqianPeng]

Hello, I encountered the same error with the languages Chinese and Japanese, I found that in udify/models/tag_decoder.py, the adaptive_cutoffs I got was [0,0] that caused the issue. So I change self.adaptive as false if the self.num_classes is small, it seems that this solves my problem.