Demo for this model on Spaces

#1
by multimodalart - opened

Hi @n4ze3m 🤗

I'm Apolinario, from the open-source team at Hugging Face. Congrats and thanks for open-sourcing n4ze3m/tiny-voca on the Hub! We were excited about this work and built with an agent an interactive demo app of it on Hugging Face Spaces, running on a free ZeroGPU infrastructure.

Here's a link to the demo: https://huggingface.co/spaces/hugging-apps/tiny-voca

We would love to transfer this demo to you or your organization. Would you like this demo to live under your own account or organization? If so just let me know here which username to transfer to, and we'll transfer the Space over to you, we hope it can give your work more visibility, discoverability and allows folks to try it out.

(If you have any questions or just want to chat more about this, you can find me on Twitter, LinkedIn or apolinario @ huggingface.co)

Cheers,

Poli

Hi Poli, thank you, this is a really nice surprise! Yes please, transfer it to n4ze3m.

I also have a small browser demo at https://tiny-voca.n4ze3m.com that runs fully on the device, but having a Space on the Hub is great for people who find the model there.

Thanks again to you and the team. 🤗

Heads-up on the Space in this thread, because it matters before the transfer: hugging-apps/tiny-voca is in RUNTIME_ERROR right now, and has been since 15 Nov. The runtime error message is:

No @spaces.GPU function detected during startup

That is the whole problem. A ZeroGPU Space starts, then the platform kills it unless at least one function in the app carries the @spaces.GPU decorator, and the app log looks healthy right up to the kill: it prints "Running on local URL" and then stops. Adding spaces to requirements and decorating the inference function is enough to get past it. The pipeline state is not recoverable from the UI, the Space has to be restarted after the fix.

I hit the same wall building mine and wrote down the rest of what ZeroGPU costs, since it matters for a model this small: a free account gets about 180 seconds of GPU per day, a @spaces.GPU(duration=10) call reserves about 15, and tiny-voca answers in 3 to 5 seconds wall time for a sentence. So the GPU is barely used here. If the goal is a permanent demo on the model page, this model would also run entirely on CPU, and the only reason it needs ZeroGPU at all is that Gradio Spaces on that hardware are what a free account can host.

Mine is at https://huggingface.co/spaces/abderrahmane11/tiny-voca if you want to compare implementations. It exposes voice, speed, pitch shift and pitch movement, shows the phoneme string, and can render both voices on one line with a single seed. No objection if it stays as the community option and the official one lands under the author's account.

Sign up or log in to comment