Cannot download tensorflow model of cahya/bert-base-indonesian-522M - bert-language-model

Question

I was going to download this model, and then I was going to save it later to be used with bert-serving. Since bert-serving only supports tensorflow model, I need to download the tensorflow one and not the PyTorch. The PyTorch model downloads just fine, but the I cannot download the tensorflow model. I used this code to download:
from transformers import BertTokenizer, TFBertModel
model_name='cahya/bert-base-indonesian-522M'
model = TFBertModel.from_pretrained(model_name)
Here's what I got when running the code on Ubuntu 16.04, python3.5, transformers==2.5.1,
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/home/username/.local/lib/python3.5/site-packages/transformers/modeling_tf_utils.py", line 346, in from_pretrained
assert os.path.isfile(resolved_archive_file), "Error retrieving file {}".format(resolved_archive_file)
File "/usr/lib/python3.5/genericpath.py", line 30, in isfile
st = os.stat(path)
TypeError: stat: can't specify None for path argument
And here's what I got when running it on Windows 10, python 3.6.5, transformers 3.1.0:
---------------------------------------------------------------------------
OSError Traceback (most recent call last)
C:\ProgramData\Anaconda3\lib\site-packages\transformers\modeling_tf_utils.py in from_pretrained(cls, pretrained_model_name_or_path, *model_args, **kwargs)
579 if resolved_archive_file is None:
--> 580 raise EnvironmentError
581 except EnvironmentError:
OSError:
During handling of the above exception, another exception occurred:
OSError Traceback (most recent call last)
<ipython-input-3-c2f14f761f05> in <module>()
3 model_name='cahya/bert-base-indonesian-522M'
4 tokenizer = BertTokenizer.from_pretrained(model_name)
----> 5 model = TFBertModel.from_pretrained(model_name)
C:\ProgramData\Anaconda3\lib\site-packages\transformers\modeling_tf_utils.py in from_pretrained(cls, pretrained_model_name_or_path, *model_args, **kwargs)
585 f"- or '{pretrained_model_name_or_path}' is the correct path to a directory containing a file named one of {TF2_WEIGHTS_NAME}, {WEIGHTS_NAME}.\n\n"
586 )
--> 587 raise EnvironmentError(msg)
588 if resolved_archive_file == archive_file:
589 logger.info("loading weights file {}".format(archive_file))
OSError: Can't load weights for 'cahya/bert-base-indonesian-522M'. Make sure that:
- 'cahya/bert-base-indonesian-522M' is a correct model identifier listed on 'https://huggingface.co/models'
- or 'cahya/bert-base-indonesian-522M' is the correct path to a directory containing a file named one of tf_model.h5, pytorch_model.bin.
This also happens with other cahya/ models. This page says that you can use the tensorflow model. However, based on the error, it seems like the file does not exist over there?
I tried downloading other pretrained model like bert-base-uncased etc. and they download just fine. This issue only happens with cahya/ models.
Am I missing something? or should I report this issue to forum or the github issue?

Answer 1

This seems to be purely an issue of your environment.
Running the first code sample worked fine for me under Ubuntu 18.04 (I think using at least Ubuntu 16.04 should work as well, Windows 10 I cannot guarantee). I further use transformers 3.1.0, and tensorflow 2.3.0.
The first environment to me seems to be purely the fault of outdated versions for both Python (recommendation in general is at least 3.6+ currently, not even tied to transformers specifically), as well as the latest transformers release for full compatibility with models from the ModelHub.
For the second enviornment, I cannot full confirm this, but I supsect that it is due to path handling under Windows 10, as transformers needs to interpret paths as either an OS path, or ModelHub id.