XXXXRT666

Results 297 comments of XXXXRT666

pull the latest code and set browser to dark mode

Decoding just one token should produce nothing, or at most a very short piece of audio

如果不是torch的问题的话那应该就是这里吧https://github.com/RVC-Boss/GPT-SoVITS/blob/13573a1b06515973de5de7435d5cc3881bc575d4/GPT_SoVITS/AR/modules/patched_mha_with_cache.py#L41 这里有未定义的Tensor和Tuple

整个文件都是,把上面的注释取消注释就行

upgrade pytorch or downgrade mkl

> i was getting this with torch2.5 > `RuntimeError: cuDNN Frontend error: [cudnn_frontend] Error: No execution plans support the graph.` > This fixed it for me. > `pip install "torch

无切分要求,最长支持近一分钟音频,只是你的显存不够

治标不治本,多worker提升利用率只是因为是CPU Bound,事实上多Worker是最低效的方式

当前webUI支持手动选取语言 `python webui.py `