Spaces:
Paused
GPU task abborted error when generating audio
This app is a joy to use, but unfortunately there is a GPU task abortion which keeps happening when I try to generate audio. Any tips would be so helpful, thank you!
There is now a new error; there is a runtime error when attempting to generate speech. I am using reference audio of roughly 10 seconds in mp3 format and have kept my prompts to about two sentences. Should I be using different files or file types? Thanks for your help and for the update today!
Hi, thanks for the report. I've been away for a while.
The space should be working now.
This happened because Huggingface upgraded the zerogpu spaces to sm_120 (Blackwell), and the turbomind backend doesn't support it.
So I've swapped it to pytorch (and pinned it to 2.8 for ncodec).
Let me know if it breaks again.
Hi there! I am so thankful for your kindness and your explanation; I think pinning pytorch to 2.8 as well as the other fixes have done the trick. It only breaks when I write too much text, but I can indeed confirm it is working and very high quality, very high quality. I really appreciate this, and I will flag any issues if I run into them. Cheers!