assert self.queue_size % batch_size == 0 # for simplicity

[scarydemon2]

我自己在进行blip_pretrain的时候，模型会经常在几百个step之后报告assert self.queue_size % batch_size == 0 # for simplicity错误，我想知道哪种情况可能会触发这种错误。我使用的是distributeddataparallel，在4个gpu上4进程训练

When I am doing blip_pretrain myself, the model will often report
“ assert self.queue_size % batch_size == 0 # for simplicity ”error after a few hundred steps,
I want to know which situation may trigger this error.
I am using distributeddataparallel with 4 process training on 4 gpus

[scarydemon2]

我发现

[EuterpeK]

请问是怎么解决的呢？