In this update, we have significantly improved character loading times by integrating a parallel processing library OpenBLAS.
OpenBLAS is an optimized BLAS (Basic Linear Algebra Subprograms) library based on GotoBLAS2 1.13 BSD version. See: https://github.com/OpenMathLib/OpenBLAS
Since neural networks are ALL linear algebra, by utilising optimised algorithms that take advantage of CPU multi-processors, we are able to improve both character processing times and response times.
For big characters (characters with long prompts), it has been observed to improve loading times by more than 30% on a Samsung S23 Ultra Snapdragon CPU.
Full Changelog
New features:
significantly improved character loading and long message processing times by integrating OpenBLAS, a CPU parallel processing library
New app: Piper Text to Voice - English (Great Britain) accent!
Improvements:
advanced settings screen now saves and validates settings on exit, instead of as you type
added advanced settings in character creation
Comments