train resource

[6lyc]

How many gpus should we use to train with 7B / 8B model?

[Suan-suannnn]

I used 4 RTX PRO 6000 for training the 1.7B model because the training stage1 may occupy at least 70GB GPU memory. Additionally, I spent more than 12 hours of stage1 traning on the Beauty dataset. About the stage 2, its memory occuptation is less than the first stage.

[leizton]

@Suan-suannnn stage2 RA部分也用了title和categories，为什么内存会比stage1少呢？