Why can’t training start?

[piwawa]

The epoch 0 has been going on for 2 days. I have filled in the path correctly in train.txt and text.txt, the dataset has 330k videos and has been preprocessed. If I select some data from train.txt, it can start training immediately. Why can't use a large dataset to train?

[bjfrbjx]

DataSet里的continue打一下断点，原项目写的数据加载方式总会跳过异常，变成无效死循环

[piwawa]

DataSet里的continue打一下断点，原项目写的数据加载方式总会跳过异常，变成无效死循环

Syncnet能训了，不过loss一直在0.69。

python hq_wav2lip_sam_train.py 现在训wav2lip又出问题了，用的一模一样的数据集，一直卡在这不动，4090显卡，200多G内存，卡好几天没反应。

[yuanmaitian]

DataSet里的continue打一下断点，原项目写的数据加载方式总会跳过异常，变成无效死循环

大佬，打断点是指调试？还是说直接break？我的也是，用了你的stream-wav2lip的训练syncnet，调好了index error后，一直卡在epoch 0不动。

[qiuzi]

@piwawa 老哥你是怎么解决epoch 0 问题的？