[Bug] Inference API for intfloat/multilingual-e5-large-instruct unreachable (health check failed for days)

[Chris-K-dev]

Describe the bug

Hi Hugging Face team,

The public Inference API for intfloat/multilingual-e5-large-instruct has been down for several days.
The model shows the yellow warning ("Our latest automated health check did not complete successfully"), and all POST requests to the HF Inference API return 502 errors.

This model is part of your Model Catalog and heavily used for embeddings (e.g., via n8n integrations).
It seems the backend container is inactive or stuck. Could someone please re-trigger the deployment or health check?

Repro: simple cURL call fails consistently.
Reference: https://huggingface.co/intfloat/multilingual-e5-large-instruct/discussions/30

Thanks a lot for looking into it 🙏

area:inference-api
type:bug

Reproduction

No response

Logs

System info

Not applicable — issue relates to Hugging Face Inference API server-side deployment, not local SDK.

[Wauplin]

Hi @Chris-K-dev , sorry for the inconvenience. I just tried https://huggingface.co/intfloat/multilingual-e5-large-instruct?text=hello+world and it seems to work in the widget. So it seems that the model has been restarted (either automatically or by someone). Is it all good on your side?

[Chris-K-dev]

Thanks for checking, @Wauplin — you’re right, intfloat/multilingual-e5-large-instruct is working again 👍

However, the non-instruct version (intfloat/multilingual-e5-large) is now unreachable via the Inference API.
Could you please recheck or restart that one as well?

Thanks a lot for your help 🙏

[Wauplin]

Hi @Chris-K-dev , it seems that https://huggingface.co/intfloat/multilingual-e5-large is back again. Could you confirm?

[Chris-K-dev]

Hi @Wauplin,

Thanks again for the quick support 🙏

Just wanted to mention that the model worked perfectly for a couple of days after your last update, but it’s unfortunately hardly reachable again now.

Is there any way to make these inference models behave more reliably long term?
Or could you share a bit about the underlying mechanism that causes them to occasionally become unreachable (and then recover again)?

It would help a lot to understand how to plan around it or add fallback logic on our side.

Thanks so much for your time and help!
Chris