siyuanliii/masa

Issues about batch size

Opened this issue · 1 comments

Thanks for your great work! I want to follow your work and finetune MASA-Adapter, but I only have 8 2080ti GPUs. I notice that the batch size was set to 128 in the paper. Can I use a smaller batch size? Does the batch size have a big impact on the final results?

Thanks for your interest! Yes, you can decrease the batch size. When you use a smaller batch size, please make sure to adjust your learning rate accordingly. I wouldn't expect a huge difference in the results.