显存泄露 #213
Answered
by
snakers4
speakstone
asked this question in
Q&A
显存泄露
#213
-
|
在使用本项目时，对于特别长的音频，如50min以上，model(chunk, sampling_rate)执行完后，对于下一个chunk，并不会主动释放显存，导致每个speech_prob推理需要的显存不断增加，这对于大的语音数据将直接显存溢出或cpu溢出 |
Beta Was this translation helpful? Give feedback.
Answered by
snakers4
Aug 15, 2022
Replies: 1 comment
-
|
This typically happens when you do not disable gradients in pytorch |
Beta Was this translation helpful? Give feedback.
0 replies
Answer selected by
snakers4
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
This typically happens when you do not disable gradients in pytorch