Model is not cached even though HF_HOME / HF_HUB_CACHE are set correctly

[magic merlin]

I’m experiencing an issue where the Hugging Face model cache is no longer being reused across runs, even though all relevant cache environment variables are set correctly and point to a persistent volume.

Important note:
This setup was working correctly until yesterday.
The same code, container image, and cache volume were successfully reusing the cached model.
However, starting this morning, the cache is no longer detected and the model is downloaded again on every startup, without any intentional changes on my side.

At startup, the logs now always show:

[HF CACHE] ❌ NOT cached: qwen/qwen-image-edit-2511@default [HF CACHE] Model snapshot not found in cache. It will be downloaded now (if network allowed).

[lament sparrow]

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

[quasi vault]

Same

[magic merlin]

It seems like there might be an issue in the data center, or an error occurred while RunPod was in the middle of an update.

[vast palm]

Same problem here... Multiple hours with service down because no model cached....

[magic merlin]

Since there hasn’t been any update or incident notice, I’m a bit unsure whether this issue is currently being addressed......

[vast palm]

And im being billed this "dead initialization time" relying on their model cache and all serverless stopped working. Can anyone check his issue @quiet elm ?

[quiet elm]

@magic merlin

Escalated To Zendesk

The thread has been escalated to Zendesk!

[atomic thorn]

can you please open a ticket for staffs to check this, please click that button

[quasi vault]

Still

[quasi vault]

Still

[atomic thorn]

Have you opened a ticket?

https://Contact.runpod.io

[atomic thorn]

This discord thread isn't an official ticket, it's for community help but staff might see it..

If you want staffs only to see it please open a ticket

[quasi vault]

Again?

[quasi vault]

Currently met this issue on EUR-IS-2 but EU-RO-1 is ok

[atomic thorn]

Oh you have? Nevermind then