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
2.73k stars 288 forks source link

Add Pytorch version check when export onnx #216

Closed KakaruHayate closed 3 weeks ago

KakaruHayate commented 3 weeks ago

在以前,使用Wavenet作为backbone时,onnx并不能在如pytorch2.0导出,故没有该功能的必要 现在,使用Lynxnet作为backbone时,不存在不能在高版本pytorch导出的情况 此外,很多时候在高版本pytorch导出的onnx也可以偶尔正常使用,少量测试难以发觉,但终究存在问题