VoiceprintRecognition-Pytorch icon indicating copy to clipboard operation
VoiceprintRecognition-Pytorch copied to clipboard

Published 20 hours ago • verified publisher icon yeyupiaoling

speakers数量太大，导致最后的全连接层参数量巨大，然后就CUDA out of memory了

Open quliulangle opened this issue 9 months ago • 5 comments

speakers得有100万，batchsize已经设置的很小了，还是出错

quliulangle avatar Apr 29 '24 08:04 quliulangle

@quliulangle 这么大的说话人吗？有没有把speed_perturb_3_class设置为False

yeyupiaoling avatar Apr 29 '24 08:04 yeyupiaoling

在配置文件里面设置的。打印的模型结构里面，最后的分类层多少参数？

yeyupiaoling avatar Apr 29 '24 09:04 yeyupiaoling

image

quliulangle avatar Apr 29 '24 14:04 quliulangle

是有点离谱哦，显存多大，还不够？

yeyupiaoling avatar Apr 29 '24 14:04 yeyupiaoling

@quliulangle speed_perturb_3_class设置为False

yeyupiaoling avatar Apr 29 '24 14:04 yeyupiaoling