inaccurate output

#1
by eschmidbauer - opened

I ran the script on a 16k audio file, and i only get this over and over:

image.png

Hey!

I'm so sorry about that. I'm away from home right now but will look into it later today.

In the meantime, could you please supply me with the code you used to inference, alongside the audio? If you don't feel comfortable including the latter, don't worry.

You can also try using the other set of decoders. Currently, there are two versions in the repo: One is FP32, the other is Q4. Which did you use for this run?

A repetition penalty could also help, but it'd need to be manually implemented. If you try that out, let me know how it goes.

I'll do my best to recreate the behavior and debug it.

Thank you for the patience!

Sign up or log in to comment