Re:Speech Synthesis on Linux (2020)
by codetrotter on Saturday September 25, 2021 @09:35AM
(#28653064)
Speaking of retraining, I think it's also potentially a bit hairy with regards to reproducible builds don't you think? My impression is that machine learning models are often initialized with random values before training begins. And some of them may use additional random data while training as well, I think.
I think a bigger issue is that retraining isn't deterministic (ie bitwise identical) across different hardware, IIRC due to differing float behaviour. I guess folks interested in ML reproducible builds will have to ignore small differences in model float values, but I wonder what affect those differences on model outputs.
Re:Speech Synthesis on Linux (2020)
by DiogenesKynikos on Saturday September 25, 2021 @09:52AM
(#28653208)
You can fix the seed for the pseudorandom number generator.
https://salsa.debian.org/deeplearning-team/ml-policy
Stuffing an existing model into a .deb is of course fairly easy.