How sensitive is this model to different batch size?

[yxchng]

Will small batch size like 512 work? I only have 8 GPUs.

[LTH14]

The smallest batch size I tested is 1024, which gives a similar performance. Since we have a learning rate scaling w.r.t. the batch size, I guess the performance will not degrade much with bsz=512, but I'm not very certain.

[cannonli7]

Will small batch size like 512 work? I only have 8 GPUs.

Hello， could you tell me how to reconstruct an image work with MAGE? I get the output image almost the same as the input image with the released checkpoint? could you help me with that?