Spaces:

---

zero-gpu-explorers

/

README

like 102

Running

Hi Hugging Face team,

I’m running a ZeroGPU Space cloned from the official LTX-2.5 Space.

Clone: Thananon/ltx-2-5-distilled-zero
Official reference: Lightricks/LTX-2.5

The official Space is currently RUNNING, while the clone consistently reaches RUNTIME_ERROR.

Environment
Hardware: zero-a10g
Clone revision: fdfce6d50477b5b7011c0f4b29d060265b3651b0
spaces==0.51.1
Build completes successfully.
Model snapshot download completes (~30.8 GB).
Model initialization and .to("cuda") complete under ZeroGPU CUDA emulation.
Failure

The process fails during ZeroGPU tensor packing:

OSError: [Errno 28] No space left on device

Call path:

demo.launch()
→ spaces.zero.gradio
→ ZeroGPU startup
→ torch.pack()
→ pack_tensors()
→ os.posix_fallocate(fd, 0, total_asize)

Packing reports:

0.00 / 82.1G

The failure occurs before the Space becomes operational.

We also checked spaces 0.51.3; the relevant packing behavior appears unchanged.

Question

Could you please clarify:

Is an ~82.1 GB ZeroGPU tensor pack expected to fit within the current ZeroGPU runtime environment?
Does the official LTX-2.5 Space receive different ephemeral disk/storage handling from clones?
Is there a supported way to increase ZeroGPU ephemeral disk or avoid this packing failure?
Is this a known limitation/bug in ZeroGPU packing for large models?
Is there a recommended deployment pattern for cloning the official LTX-2.5 Space successfully on ZeroGPU?

We have already verified that this is not a build-stage failure and have avoided further code changes so the failure can remain reproducible.

Thank you.