Where to find training config for trained models?

[main-2983]

I want to train PoolFormer using new timm train.py version, which in this repo is different from PoolFormer train.py

Is there any way to get your training config to reproduce the result of PoolFormer using your new train.py

[mjack3]

That would be great but I think we do not have that information (not all)

[rwightman]

@main-2983 @mjack3 they are close enough that you can work through their args / defaults and the timm ones and match it almost exactly. Does require looking through the code a bit but it works, I've done the comparisons for deit and swin which are related.