Hi @Amy_Bray - can you please make more clear this training reproduce rule - Specifically, we should be able to re-create your submission on a single-GPU machine (eg Nvidia A100) with less than 20 hours. -Does it mean that the upper bound is A100 with 20 hours? or for example, can I use 8xA100 for 20 hours? because it is a pretty different compute amount
"a single-GPU machine" would make me assume it is a single GPU (1xA100).
indeed, my bad, I missed that part, but ok H100 for example also acceptable or not? and which one A100 is acceptable? 80gb? 40gb?
@Amy_Bray so A100 is upper bound or H100 is also ok?