Spaces aren't allocating for ZeroGPU tonight

Spaces
1

Space never allocated: hardware.current: null, builds succeed, app never runs

Claude wrote this up to save some time. Been a bit of a journey tonight trying to get it working, so far no dice.

Space: AbstractPhil/alephllm-chat (Gradio SDK, ZeroGPU requested)
Account: AbstractPhil · Observed: 2026-08-14 → 2026-08-15

Symptom

The Space builds successfully on every commit and is never allocated a
machine. Runtime API reports:

stage    : APP_STARTING        (indefinitely)
hardware : {"current": null, "requested": "zero-a10g"}
errorMessage : null

current: null persists for hours. Switching the request to cpu-basic
does not allocate either, so this is not ZeroGPU-specific.

Build is not the problem

Build log completes normally every time:

--> RUN --mount=target=/tmp/requirements.txt ... CACHED
--> COPY --link --chown=1000 ./ /home/user/app   DONE
--> Pushing image                                 DONE 0.8s
--> Exporting cache                               DONE 0.5s

The app is not the problem

The identical app.py runs correctly outside Spaces (same Python 3.12,
same gradio, same spaces package): it boots, serves /config, and
answers requests. It is the canonical ZeroGPU shape — model loaded at
module level and moved to cuda, @spaces.GPU on the generator, plain
demo.launch(). No threads, no locks, no asyncio, no coroutines, no
subprocesses of our own.

Related history on this account (may be the same underlying issue)

  1. The predecessor Space (now AbstractPhil/alephllm-chat2) spent ~16h
    serving a two-day-old commit: new builds succeeded, containers
    booted cleanly with Running on local URL, but traffic kept going to
    the old revision and runtime.sha never advanced. Factory rebuild and
    restart had no effect. Recreating the Space under a new name worked
    once, then the new Space hit the current: null state described above.

  2. alephllm-chat2 is now PAUSED but still reports
    hardware.current: zero-a10g    , with requested: cpu-basic stuck in
    flight. Re-issuing the hardware change and restarting the Space does
    not apply it. It appears to be holding a hardware assignment it was
    asked to release.

Similar reports

Ask

  1. Why is no machine being allocated for AbstractPhil/alephllm-chat
    despite a successful build and a pending hardware request?
  2. Can the stuck hardware assignment on AbstractPhil/alephllm-chat2
    (current: zero-a10g while PAUSED, requested: cpu-basic) be
    released?

Additionally, what is this TWO LINKS PER POST thing for new users? I’m nowhere near a new user for huggingface.

2

Same here and there: zero-gpu-explorers/README · All ZeroGPU Spaces stuck on "Starting" with clean logs — CPU and paid GPU unaffected @hysts

3

Well on the plus side, the AlephLLM Mini-Beatrix is perfectly playable on cpu. Pretty fun ByteLM structured system turned out pretty smart.

4

Thanks for reporting this. Looks like it’s an infra issue. I’ll check with the team internally.

5

Thank you!!

6

I’m also having the same issue. Just stuck in “app starting”

7

I’m experiencing the exact same issue. My ZeroGPU Space is stuck on ‘Starting’ and never transitions to ‘Running’.

8

thanks :heart_eyes: for this

9

yes same :zero-gpu-explorers/README · All ZeroGPU Spaces stuck on "Starting" with clean logs — CPU and paid GPU unaffected

10

Should be resolved now. Thanks!

12

I confirmed! Thanks a lot!