Spaces:

---

zero-gpu-explorers

/

README

like 102

Running

Since around 00:45 UTC on 15 August 2026, every ZeroGPU Space on my account has been stuck on "Starting" and never transitions to "Running". The container logs are completely clean.
What the logs show
The app initialises fully, downloads its assets, reports itself listening, and then nothing further — no errors, no traceback, no further output:
[notice] A new release of pip is available: 26.1.2 -> 26.2.1
[notice] To update, run: pip install --upgrade pip
GRADIO_HOT_RELOAD: Launching demo not found in main. Using 'demo'

• Running on local URL: http://0.0.0.0:7860
• To create a public link, set share=True in launch().
  No requests ever reach the app, so the readiness probe appears never to succeed.
  What I have ruled out
  Not the application. Another ZeroGPU Space on the same account — a portfolio site containing no model, no inference and no GPU code at all — is equally stuck. Both fail; both have clean logs.
  Not the account or the platform generally. My CPU Space and my paid-GPU Space both start normally, right now, on the same account.
  Not SSR. Set GRADIO_SSR_MODE=False; confirmed by the absence of with SSR ⚡ in the log. No change.
  Not startup work. Moved all heavy initialisation (a git clone, a 562 MB checkpoint download, a pip install) out of import time so the server binds within seconds. No change.
  Not CUDA in the main process. Removed a module-level model preload that was initialising CUDA outside @spaces.GPU. No change.
  Factory rebuild, several restarts.
  Environment
  SDK: gradio, sdk_version: 6.19.0
  Hardware: ZeroGPU
  spaces==0.51.1 (was 0.51.0 when the Space last started successfully, on 31 July)
  Account: PRO
  The pattern matches the June 2026 incident where Spaces reached Running on local URL but the status never advanced, which turned out to be platform-side. This time it appears confined to ZeroGPU.
  Is anyone else seeing this today?