How much GPUs are needed to Finetune SigLIP+Stable Diffusion?

[CnFaker]

Excellent work!
What dataset did you use to train SigLIP + Stable Diffusion, how many GPUs were used, and how long did the training take?

[tsb0601]

Hi,

Thanks for interest in our work! It took around 4 nodes and 2 days. It roughly trained for around 10 million images. I also tried just 1 node of 8 GPUs and it works also well with less batch sizes

[CnFaker]

Thank you for your reply, it was very enlightening for me.