OSError: It looks like the config file at 'roberta-base/config.json' is not a valid JSON file

[Ayush]

I have been trying to deploy my Question Answering model on St Cloud.
As per steps, I pushed the local model contents to github with my folder name as ‘roberta-base’ which has the .bin model and json.config files.

It runs fine on my local machine but on streamlit it throws error during inference.
Repo: Ayush1702/ESG-Question-Answering (github.com)

@st.cache(allow_output_mutation=True)
def question_answering():
    model_name = "/app/question-answering/roberta-base/"
    model = AutoModelForQuestionAnswering.from_pretrained(model_name)
    return model

if st.button('Submit'):
    context_input = st.session_state.context
    question_input = st.session_state.question_default
    with st.spinner('Loading Model'):
        my_model = question_answering()
    tokenizer_path = "/app/question-answering/roberta-base/"
    tokenizer = AutoTokenizer.from_pretrained(tokenizer_path)
    question_answerer = pipeline("question-answering", model=my_model, tokenizer=tokenizer)
    result = question_answerer(question=question_input, context=context_input)
    st.write(HTML_WRAPPER.format(result['answer']), unsafe_allow_html=True)

>>> OSError: It looks like the config file at '/app/question-answering/roberta-base/config.json' is not a valid JSON file.

[snehankekre]

Hi @Ayush

Could you please link to your GitHub repo so we can help debug?

[Ayush]

Here

[snehankekre]

I’d firstly suggest using relative references instead of passing a GitHub URL as a file path. URLs are not valid file paths.

Before

model_name = "https://github.com/Ayush1702/ESG-Question-Answering/blob/main/roberta-base/"
tokenizer_path = "/app/esg-question-answering/roberta-base/"

After

model_name = "roberta-base/"
tokenizer_path = "roberta-base/"

Next, you likely need to fix your config.json file. Specifically, the value of _name_or_path key, whose current (and possibly incorrect) value is https://github.com/Ayush1702/ESG-Question-Answering/blob/main/roberta-base/.

[snehankekre]

If after the above changes, you’re still running into the error, it may be an issue with Git LFS as described here:

OSError: Unable to open file (from githib)

Hi @Taiwo_Osunrinde I suspect the issue is with Git LFS. When I clone your repo and checked the model file size, it shows up as only 4Kb instead of 176 MB This might happen when a user has maxed out their Git LFS bandwidth or storage limits. Would you mind following these instructions from GitHub to view your Git LFS storage and bandwidth usage? Does it indicate that you have exceeded the default 1 GB of storage and/or bandwidth? Solution Take a look my fork of your app. S…

In that case, you would have to store your model elsewhere (e.g. Dropbox, AWS, GCP) and download and cache it from your app.

[Ayush]

Modified Paths yet no change, config file is fine too
Git LFS is not an issue, it is <1GB limit as model size is ~500MB

I’m not sure what else to troubleshoot, Can you please check by cloning?

[snehankekre]

I’ve raised this issue with our team. It may or may not be an issue with Git LFS, we’re not certain

OSError: It looks like the config file at 'roberta-base/config.json' is not a valid JSON file

If after the above changes, you’re still running into the error, it may be an issue with Git LFS as described here: In that case, you would have to store your model elsewhere (e.g. Dropbox, AWS, GCP) and download and cache it from your app.

[Ayush]

Hi Snehan,

Any luck?
Can you try checking with Python 3.7, I’m unable to change python version from settings.

[snehankekre]

The Python version is likely not an issue. To change Python versions, you will have to delete the app, and select the appropriate Python version from the Advanced settings modal before you re-deploy your app:

Deploy an app - Streamlit Docs

We still don’t have an update on the GitLFS issue.