Grants

The flow

1. Build the Space. If the user still has a free ZeroGPU slot, create it as --flavor zero-a10g and iterate normally with real inference before applying.

   If they’re out of slots, create a Static Space instead (--space-sdk static — free for everyone) and push the app there; the SDK can be switched to gradio in the README frontmatter once the grant lands. Code the app for ZeroGPU anyway — import spaces, @spaces.GPU, module-scope .to("cuda"). In this mode you can’t iterate-with-real-inference before the grant, so just get the code in place and submit.

   For a dedicated-GPU grant, get the app to BUILD cleanly and reach RUNNING (even if the runtime would OOM on real input), then submit.

2. Submit a Community Tab discussion on the Space. Title:

   Copy

   Apply for a GPU community grant: <Personal|Company|Academic> project

   Pick the closest fit. Body:

   Copy

   Description of the app: one paragraph on what it does + who it's for.
   Justification: one paragraph on why this should run on ZeroGPU
   (open-source, research, educational, etc.). 

   If the user didn’t give you a justification, a reasonable default is “Public open-source demo, can’t cover the hardware cost — happy to provide more context if helpful.”

3. Wait. Open and publicly-facing applications by researchers, tinkerers, and institutions are typically approved. Approval can take days.

4. Once approved, the hardware is attached automatically — no code change needed (a Static holding Space still needs its sdk: flipped to gradio). The user comes back and you can iterate / refine with real GPU access.

When to suggest this

• The use case is a clear public ML demo (not a private tool) and the user is out of free ZeroGPU slots.
• The model needs more than ZeroGPU offers — beyond 48 GB large / 96 GB xlarge, or a non-PyTorch runtime — and the user can’t pay.

When NOT to suggest this

• The user is on a free account and this is their 1st or 2nd Space — they can create it on ZeroGPU directly; no grant needed.
• Private / commercial / closed-source projects — push the user toward PRO instead.
• canPay=True users who just need paid hardware — they can attach it directly.

Posting the request programmatically

The Community Tab must be enabled on the Space (default — keep it on).