17 ZeroGPU Slots Reported but Only 8 Spaces Listed

Spaces
1

I was trying to run a Zero GPU space, but I reached a quota error: “Quota exceeded for flavor zero-a10g (requested=1): current=17, limit=10”.

17 zerogpu spaces assigned.



api.get_space_runtime(repo_id="r3gm/wan2-2-fp8da-aoti-preview-r")

SpaceRuntime(stage='PAUSED', hardware=None, requested_hardware='zero-a10g', sleep_time=172800, storage=None, dev_mode=False, hot_reloading=None, volumes=None, raw={'stage': 'PAUSED', 'hardware': {'current': None, 'requested': 'zero-a10g'}, 'gcTimeout': 172800, 'errorMessage': 'Quota exceeded for flavor zero-a10g (requested=1): current=17, limit=10', 'replicas': {'requested': 1}, 'devMode': False, 'domains': [{'domain': 'r3gm-wan2-2-fp8da-aoti-preview-r.hf.space', 'stage': 'READY'}]})

However, when I verify my spaces that request the zero-a10g hardware, I only see 8 spaces, and 7 of them are running.

ID                                  Stage           Replicas

r3gm/wan2-2-fp8da-aoti-preview      RUNNING         {'current': 6, 'requested': 'auto'}
r3gm/AICoverGen                     RUNNING         {'current': 1, 'requested': 1}
r3gm/rvc_zero                       RUNNING         {'current': 1, 'requested': 1}
r3gm/Audio_separator                RUNNING         {'current': 1, 'requested': 1}
r3gm/DiffuseCraft                   RUNNING         {'current': 1, 'requested': 'auto'}
r3gm/upscale_board                  RUNNING         {'current': 1, 'requested': 1}
r3gm/wan2-2-fp8da-aoti-old-backup   RUNNING         {'current': 1, 'requested': 1}
r3gm/wan2-2-fp8da-aoti-preview-r    PAUSED          {'requested': 1}

The only explanation I can think of is that replicas are also counted in the Zero GPU slot limit. If two spaces are set to “auto” replicas, their maximum of 6 each would add up to 12, and when combined with the other running spaces, the total becomes 17.

2

Hi @r3gm , thanks for flagging this. I’ll check with the infra team internally.

3

It does look like replicas are being counted toward the Zero GPU quota. Two auto Spaces (up to 6 replicas each) plus your other Spaces add up to 17, which matches the error. If reducing replicas doesn’t help, it’s likely a backend quota issue and worth contacting Hugging Face support.

4

Hi @r3gm, I’m looking into it. Replicas shouldn’t count toward the limit, and it will be fixed as soon as possible if it turns out this is the case. Thanks for reporting it!

5

@r3gm this should now be fixed. Feel free to let me know if you still encounter issues