What GPUs were used for post-training?

[shuchenYee]

Hi! Thanks for releasing this amazing work. What GPUs and the number of GPUs were used for post training? Thanks!

[Zeng-Jia]

We conducted post-training on 8 A100 GPUs with a batch size of 8*16, consuming approximately 65-70GB of memory per GPU.