Spaces:

Nick088

/

Audio-SR

like 24

Running on Zero

Wdym what's the error? I just factory rebuilded the space and tested it myself and it works all fine, its normal if its very slow, its becuase its running on CPU, you would need to duplicate it and run it on Paid GPU

I tried the rebuilt version again with a WAV of a Rhythm Tengoku prologue sampled at 13379hz (because I'm broke and also too lazy to duplicate it) and I think the basic model is a bit broken because it threw an error at me after a few seconds

pretty weird, prolly its because this space needs to be on duplicated on GPU

I duplicated the space and ran it on a few good paid GPUs & it still errors out... 🙈

Please share a successful config!

The space is now running on ZeroGPU (A100) btw, what's the error? Be sure you aren't using a long audio, as AudioSR isn't really made for long audios, if its long i would suggest splitting it into shorter ones

Thank you for switching the cpu!

Unfortunately, it's still giving me an error even when I use a 5 second sample from the github.
Just says error.

Thank you for switching the cpu!

Unfortunately, it's still giving me an error even when I use a 5 second sample from the github.
Just says error.

Show the inputs (dim steps, audio file, etc) and the output logs too so I can analyze and see how to help you to fix it

I think there are typos (torch.cuda.is_avaible) in https://huggingface.co/spaces/Nick088/Audio-SR/blob/main/app.py.

Also, when running on ZeroGPU, I think there is no need to do gc or empty CUDA memory.

Just a reminder: only the space owner can see the log of space.

I think there are typos (torch.cuda.is_avaible) in https://huggingface.co/spaces/Nick088/Audio-SR/blob/main/app.py.

Thank you for telling, just fixed it

Also, when running on ZeroGPU, I think there is no need to do gc or empty CUDA memory.

Just a reminder: only the space owner can see the log of space.

I know only the space owner can see the logs, also I used GC and empty CUDA memory because audio sr requires much VRAM