First run works well, 2nd run is extremely slow - any steps "to reset" needed between ?

[Chipstix24]

First run using 15second sample, 0.7 speed multiplier and "4 minutes" worth of text, takes around 30mins (my laptop is only M1 Macbook with in built GPU, so not very powerful, but I can live with 30mins for hobby use)

Second run, I just replace the text, and leave all else the same and it has been 8hrs to get to 44% on 8 of 18 steps??

I am running e2-f5-tts in Pinokio. I noticed on the faster run, GPU in activity monitor was 80%+, On the slower 2nd run, it was around 20%. So, not sure its using the GPU properly?

Any tips appreciated!

[Chipstix24]

Update: It appears that rebooting my laptop and starting a new session of Pinokio gets the GPU % back up again and speed improves back to what it was. Maybe the laptop restart was unnecessary, but certainly a new Pinokio and e2-f5-tts session is needed each time for some reason?

[Chipstix24]

I spoke too soon. I get this error message. is there any way around this with my hardware or do I need to somehow buy a new computer!? :-)

pinokio/api/e2-f5-tts.git/app/env/lib/python3.10/site-packages/vocos/spectral_ops.py:46: UserWarning: The operator 'aten::unfold_backward' is not currently supported on the MPS backend and will fall back to run on the CPU. This may have performance implications. (Triggered internally at /Users/runner/work/pytorch/pytorch/pytorch/aten/src/ATen/mps/MPSFallback.mm:13.)
return torch.istft(spec, self.n_fft, self.hop_length, self.win_length, self.window, center=True)