Serverless Inference API error on new model

Inference Endpoints on the Hub
1

When I tested the Serverless Inference API with my new model, I encountered the following error:

{'error': "We couldn't connect to 'https://huggingface.co' to load this file, couldn't find it in the cached files and it looks like Saripudin/autotrain-model-datasaur-ZDAzZTc5NmI-NGJmYmVlOWU is not the path to a directory containing a file named config.json.\nCheckout your internet connection or see how to run the library in offline mode at 'https://huggingface.co/docs/transformers/installation#offline-mode'."}

Does this mean that Hugging Face has now limited their Serverless API usage?

2

Currently, HF’s Serverless Inference API is in the midst of a specification change and a massive error coming at the same time.
One of the errors has been fixed.

1 Like
3

Thank you @John6666

I’ve tested it and it has been worked normally

4

I have seen the same errors. Has it been fixed for you?

5

When I tested in 2 days ago, it was working normally. But it turns out It happen error again now…

6

There were generally three problems that occurred with HF this time, limited to the image generation AI,

  1. Errors that prevented the use of the Serverless Inference API, regardless of the client (can be fixed manually by following the instructions below, and new occurrences are almost never seen).
  2. An error that caused very strange server-side responses on some Gradio 3.x systems (still ongoing, no workaround, but spaces that have not been restarted after the problem are safe).
  3. Multiple fatal errors in Zero GPU space (the biggest one was fixed a few days ago but still remains)

Workaround

Setting fastapi==0.112.2 in requirements has helped me as a temporary fix.

We just released gradio==4.43 with the same fix. So new Gradio Spaces should work fine now (older ones will need to update the sdk_version in the README.md file)

Workaround in Zero GPU Space

Long shit report by me.