Inference

[ghost]

The results of inference are not the same with the same config!

[lifeiteng]

@primepake What do you mean?

[cwjacklin]

inference is quite sensitive to accuracy of --text-prompts. I find that if i add or remove punctuation such as ",", sometimes I get WARNING [words_mismatch.py:88] words count mismatch on 100.0% of the lines (1/1) or worse I get RuntimeError: Unable to find a valid cuDNN algorithm to run convolution

[cwjacklin]

prompt audio bleedding/mixing into output wav issue.

if prompt text doesn't end with a ".", end of the prompt audio will appear and get mixed into the beginning of output wav.

easily fixed though, just add a "." to prompt text at the end