Loading TTS xtts model from tts_models/multilingual/multi-dataset/xtts_v2...

[yikexiaoxing]

在这一步为什么特别慢，我设置了使用GPU加速，但还是特别慢。而且第二次使用时，依旧从新开始加载

[ROBERT-MCDOWELL]

for further comments, please translate your question in ENGLISH. thanks.

Loading an model is NOT linked to CPU or GPU, but RAM. if you have a slow or busy RAM, an xtts model is around 2GB so you can imagine how slow it can be. Also, if it's the first time you load it, it will load from huggingface platform, so they can be slow too.
when the TTS model is in your cache (./models) so it will be faster. in the coming v2.1.0 we are using fastapi module which will accelerate the process.

[yikexiaoxing]

Thanks for your answer, I tried several times and have loaded the model cache locally on huggingface, and this situation still occurs. In each generation, it does not become faster because of the first load, because the memory usage is observed to be getting larger and larger

[ROBERT-MCDOWELL]

OS? CPU? GPU? how can I give you an advice if you don't provide all hardware and software informations?
also are you using docker or not?

[yikexiaoxing]

Linux CPU 80G GPU RTX4090 and using Docker

[ROBERT-MCDOWELL]

@DrewThomasson will be able to explain to you what's happening, I saw your issue somewhere already....

[ROBERT-MCDOWELL]

this should be fixed on the last git and release. thanks to try and report if still an issue.