OSError: Can't load config for 'bert-base-uncased

[WinMinTun]

Environment info

It happens in local machine, Colab, and my colleagues also.

• transformers version:
• Platform: Window, Colab
• Python version: 3.7
• PyTorch version (GPU?): 1.8.1 (GPU yes)
• Tensorflow version (GPU?):
• Using GPU in script?: Yes
• Using distributed or parallel set-up in script?: No

Who can help

@LysandreJik It is to do with 'bert-base-uncased'

Information

Hi, I m having this error suddenly this afternoon. It was all okay before for days. It happens in local machine, Colab and also to my colleagues. I can access this file in browser https://huggingface.co/bert-base-uncased/resolve/main/config.json no problem. Btw, I m from Singapore. Any urgent help will be appreciated because I m rushing some project and stuck there.

Thanks

403 Client Error: Forbidden for url: https://huggingface.co/bert-base-uncased/resolve/main/config.json

HTTPError Traceback (most recent call last)
/usr/local/lib/python3.7/dist-packages/transformers/configuration_utils.py in get_config_dict(cls, pretrained_model_name_or_path, **kwargs)
505 use_auth_token=use_auth_token,
--> 506 user_agent=user_agent,
507 )

6 frames
HTTPError: 403 Client Error: Forbidden for url: https://huggingface.co/bert-base-uncased/resolve/main/config.json
During handling of the above exception, another exception occurred:

OSError Traceback (most recent call last)
/usr/local/lib/python3.7/dist-packages/transformers/configuration_utils.py in get_config_dict(cls, pretrained_model_name_or_path, **kwargs)
516 f"- or '{pretrained_model_name_or_path}' is the correct path to a directory containing a {CONFIG_NAME} file\n\n"
517 )
--> 518 raise EnvironmentError(msg)
519
520 except json.JSONDecodeError:

OSError: Can't load config for 'bert-base-uncased'. Make sure that:

• 'bert-base-uncased' is a correct model identifier listed on 'https://huggingface.co/models'

• or 'bert-base-uncased' is the correct path to a directory containing a config.json file

[sgugger]

Was it just a fluke or is the issue still happening? On Colab I have no problem downloading that model.

[WinMinTun]

@sgugger Hi it is still happening now. Not just me, many people I know of. I can access the config file from browser, but not through the code. Thanks

[WinMinTun]

Still not okay online, but I managed to do it locally

git clone https://huggingface.co/bert-base-uncased

#model = AutoModelWithHeads.from_pretrained("bert-base-uncased")
model = AutoModelWithHeads.from_pretrained(BERT_LOCAL_PATH, local_files_only=True)

#tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
tokenizer = AutoTokenizer.from_pretrained(BERT_LOCAL_PATH, local_files_only=True)

adapter_name = model2.load_adapter(localpath, config=config, model_name=BERT_LOCAL_PATH)

[julien-c]

This, like #12940, is probably related to a change we've made on the infra side (cc @n1t0), which we'll partially revert. Please let us know if this still occurs.

[n1t0]

@WinMinTun Could you share a small collab that reproduces the bug? I'd like to have a look at it.

[jason-weddington]

With additional testing, I've found that this issue only occurs with adapter-tranformers, the AdapterHub.ml modified version of the transformers module. With the HuggingFace module, we can pull pretrained weights without issue.

Using adapter-transformers this is now working again from Google Colab, but is still failing locally and from servers running in AWS. Interestingly, with adapter-transformers I get a 403 even if I try to load a nonexistent model (e.g. fake-model-that-should-fail). I would expect this to fail with a 401, as there is no corresponding config.json on huggingface.co. The fact that it fails with a 403 seems to indicate that something in front of the web host is rejecting the request before the web host has a change to respond with a not found error.

[julien-c]

Thanks so much @jason-weddington. This will help us pinpoint the issue. (@n1t0 @Pierrci)

[lrodrigues-at-navteca]

I have the same problem, but it only happens when the model is private.

[sgugger]

Your token for use_auth_token is not the same as your API token. The easiest way to get it is to login with !huggingface-cli login and then just pass use_auth_token=True.

[lrodrigues-at-navteca]

I think the problem is something else:

[bennimmo]

Yes, I have come across this as well. I have tracked it down to this line

transformers/src/transformers/pipelines/__init__.py

Line 422 in 1437382

model_kwargs["use_auth_token"] = model_kwargs.get("use_auth_token", use_auth_token)

It's because the use_auth_token has not been set up early enough in the model_kwargs. The line referenced above needs to be moved above instantiate config section.

[bennimmo]

I've added a pull request to which I think will fix this issue. You can get round it for now by adding use_auth_token to the model_kwargs param when creating a pipeline e.g.:
pipeline('zero-shot-classification', model=model, tokenizer=tokenizer, model_kwargs={'use_auth_token': True})

[hemangjoshi37a]

Still getting the same error
Here is my code :

from transformers import AutoModelForTokenClassification, AutoTokenizer
model = AutoModelForTokenClassification.from_pretrained("hemangjoshi37a/autotrain-ratnakar_600_sample_test-1427753567", use_auth_token=True)
tokenizer = AutoTokenizer.from_pretrained("hemangjoshi37a/autotrain-ratnakar_600_sample_test-1427753567", use_auth_token=True)
inputs = tokenizer("I love AutoTrain", return_tensors="pt")
outputs = model(**inputs)

Error :

----> 3 model = AutoModelForTokenClassification.from_pretrained("hemangjoshi37a/autotrain-ratnakar_600_sample_test-1427753567", use_auth_token=True)
OSError: Can't load config for 'hemangjoshi37a/autotrain-ratnakar_600_sample_test-1427753567'. If you were trying to load it from 'https://huggingface.co/models', make sure you don't have a local directory with the same name. Otherwise, make sure 'hemangjoshi37a/autotrain-ratnakar_600_sample_test-1427753567' is the correct path to a directory containing a config.json file

I have transformers version : 4.21.3
https://hjlabs.in

[hemangjoshi37a]

runnign this command and authenticating it solved issue: huggingface-cli login
https://hjlabs.in

[PaskalSunari]

I am facing the same problem in Kaggle too... How can I

resolve this issue ?

[lyk0013]

Hello, I had the same problem when using transformers - pipeline in the aws-sagemaker notebook.

I started to think it was the version or the network problem. But, after some local tests, this guess is wrong. So, I just debug the source code. I find that:

This will raise any error as EnviromentError. So, from experience, I solve it, by running this pip:
!pip install --upgrade jupyter
!pip install --upgrade ipywidgets

You guys can try it when meeting the problem in aws-notebook or colab!

[VRDJ]

I am unable to solve this issues Since Morning .. i had been trying to Solve it ...

Im working on my Final Year Project .. can someone pls help me in it ...

[hemangjoshi37a]

Just ask chatGPT LOL...😂😂

[VRDJ]

I dont understand it ?? What do u mean ..
The Hugging Face Website is also not working ...

[hemangjoshi37a]

@VRDJ goto this website chatGPT and enter your error in the chatbox in this website and for the 99% you will get your solution there.

[DuanXiaoyue-LittleMoon]

Still not okay online, but I managed to do it locally

git clone https://huggingface.co/bert-base-uncased

#model = AutoModelWithHeads.from_pretrained("bert-base-uncased") model = AutoModelWithHeads.from_pretrained(BERT_LOCAL_PATH, local_files_only=True)

#tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased") tokenizer = AutoTokenizer.from_pretrained(BERT_LOCAL_PATH, local_files_only=True)

adapter_name = model2.load_adapter(localpath, config=config, model_name=BERT_LOCAL_PATH)

---

Hello! Thanks for your sharing. I wonder in
'tokenizer = AutoTokenizer.from_pretrained(BERT_LOCAL_PATH, local_files_only=True)',
which file does 'BERT_LOCAL_PATH' refer to specifically? Is it the path for the directory 'bert-base-uncased', or the 'pytorch_model.bin', or something else?