Hi, how is it going? I used pytorch to implement fftnet as well, quite the same structure as yours with small modification.
I'm currently having problem about number of parameters. The author said that fftnet has less than 1M parameters, but my model has more than 2M parameters. Seems like fftnet is smaller than what we expect, but I can't figure out what makes it small and run fast. Do you have any idea about this?
The current generate speed is about 300 samples/s on a i7 cpu.
Hi, how is it going? I used pytorch to implement fftnet as well, quite the same structure as yours with small modification. I'm currently having problem about number of parameters. The author said that fftnet has less than 1M parameters, but my model has more than 2M parameters. Seems like fftnet is smaller than what we expect, but I can't figure out what makes it small and run fast. Do you have any idea about this? The current generate speed is about 300 samples/s on a i7 cpu.