Ross Wightman

Thanks for the work but there is no point in switching v2 for the sake of it, there will be issues with timms own transforms. I will likely keep legacy...

@ChengpengChen thanks, once I'm done/satisified with the remaining mobilenetv4 model trainings I'll see how these fair with comparable hparams. FYI, if there is any reparameterization scheme to be deployed here,...

@ChengpengChen started training a large @ 384x384 with exact hparams of the mnv4 conv-large I trained.

@ChengpengChen got sidelined by a bit of time off the other week... so I did finish training the large, but it's ~ .5% or so behind the mnv4 equivalent, so...

See https://github.com/huggingface/pytorch-image-models/discussions/2345 ... working on it along with object detection support

@eqy there's a small issue here, which is a bit to explain which is why this sat... Validation script should work in streaming mode without a defined length, I believe...

@nitin-dominic if the model is pushed with `label_names` and optionally `label_descriptions` and those are in the `config.json` it will work with the Hub API inference but yeah, `inference.py` is not...

@yukke42 did you actually try it? python 3.12 and the latest pytorch 2.5 works fine for me .. setup metadata might be out of date but everything works

@artemisart thanks for looking at this, I gave up long ago due to how fiddly these issues are, heh. So, there is a `torch.jit.is_tracing()` but I haven't used it much....

I don't think there's anything I can do here, and it's likely transient across torch / numpy / python versions so closing