Millennium Whisper · published · build 19431883
We hope you're enjoying all the goodies from v0.2! v0.2.1 has come around fast as we've tried to fix a bunch of minor bugs and adjustments but also some unforeseen major issues with hardware compatibility.
As mentioned in the v0.2 release notes, we've shifted our AI over to a new back-end inference engine. This was to facilitate voice input, cross-platform compatibility but also a bunch of exciting AI improvements in future.
What we didn't anticipate for this new engine to not work on a bunch of people's hardware... sorry about that.
If you were one of those effected you can now switch back to the legacy engine by selecting DML in the "model type" drop-down in the options menu. This will only work on Windows and you cannot use voice input! We'll try and transfer as many AI improvements as we can to the legacy engine but pretty soon we're going to be unable to. We've been looking into alternatives so stay tuned. If your GPU still doesn't work try the CUDA option to see if that fixes it.
Many have also said the new engine is slower. We're aware of this and are working on optimisations to bring it more in line with (and hopefully eventually overtake) the legacy engine. We've added a timer to the bottom of the options menu to more easily do a comparison.