ImportError with Gradio

[MiguelHGDC]

try:
    import gradio as gr
except: 
    !pip -q install gradio
    import gradio as gr
    
print(f"Gradio version: {gr.__version__}")

Output: ImportError: cannot import name 'deprecated' from 'typing_extensions'

My versions:

python version: 3.8.10
torch: '2.0.1+cu118'
torchvision: '0.15.2+cu118'
typing_extensions: 4.7.1
gradio==3.44.3

Does anyone know what the compatibility problem is?

[MariaKrakowiak]

Hi @MiguelHGDC ,

Did you manage to solve above issue? I have similar one.

[mrdbourke]

Hi @MiguelHGDC and @MariaKrakowiak ,

Are you still able to use Gradio?

Are you using Google Colab or a local machine?

I just tried to install/import it in Google Colab and it's showing a similar error but the conflict is with tensorflow-probability (not needed for the course or to use Gradio.

See the notebook here: https://colab.research.google.com/drive/1f2-ptPGecjPQ0zMNnS6rTVyHmR8i6izv?usp=sharing

However, this conflict does not influence the use of Gradio.

[mrdbourke]

Also, perhaps this thread will help: #699 (similar issue)

[MariaKrakowiak]

Hi @mrdbourke ,

Thanks for your advice. In completely new Colab it works but unfortunately in my case it doesn't.

The thing is, I am unable to use Gradio. Before installing it, I have checked the typing_extensions version (!pip show typing_extensions) which is problematic and I received 4.5.0.

Then, I have installed Gradio via !pip install gradio with no errors and the typing_extensions version showed 4.8.0.

Finally, I tried to import the module but I have received:

In Github dedicated for this extension (https://github.com/python/typing_extensions/blob/main/CHANGELOG.md) there is information that Doc is available from version 4.8.0 which I have already installed.

Do you have any idea how to proceed?

I tried restarting and reinstalling runtime few times with no result. I am running out of ideas.

[MiguelHGDC]

I remember trying to fix this error, and I encountered the same issue as you, @MariaKrakowiak. I couldn't find the correct versions; I'm sorry I didn't respond.

Locally, it's true that it's not working... @mrdbourke, perhaps the correct solution would be to create different virtual environments for each task, especially one specifically for using Gradio.

[MariaKrakowiak]

Hi @MiguelHGDC ,

No problem - I finally overcame the issue.

My solution was to import the gradio at the beginning of Colab notebook by:

!pip install gradio
import gradio as gr

Thanks to this there were no issues with other extensions versions.

Sometimes the simplest solutions are the best 👍