GaoLeiA
Results
1
comments of
GaoLeiA
我加了个打印 def inference(self, audio_path, out_vid_name, fps, skip_save_images): os.makedirs(self.avatar_path + '/tmp', exist_ok=True) print("start inference") print(torch.is_grad_enabled())  确实为True,修改后内存大大减少,占用显存12G以下,跟LiveTalk相符。