openvpi / DiffSinger

An advanced singing voice synthesis system with high fidelity, expressiveness, controllability and flexibility based on DiffSinger: Singing Voice Synthesis via Shallow Diffusion Mechanism
Apache License 2.0
2.62k stars 275 forks source link

Better than PL auto strategy selection #159

Closed hrukalive closed 5 months ago

hrukalive commented 6 months ago

Support kwargs override for strategies even when it is set to auto. Also fixes undefined behavior (specifying single_device directly using string) in the PL code itself.

Note: this code snippet requires lightning>=2.0.