Closed lendot closed 6 months ago
The coqui engine's stock python subprocessing can cause problems when used with software that makes CUDA calls elsewhere (e.g. xtts_api_server). This fix switches from the multiprocessing module over to torch.multiprocessing.
Thanks a lot for pointing this out, will do some tests with this approach soon. This looks very promising.
The coqui engine's stock python subprocessing can cause problems when used with software that makes CUDA calls elsewhere (e.g. xtts_api_server). This fix switches from the multiprocessing module over to torch.multiprocessing.