Hi ! Thanks for the excellent work! I am trying to use the Constraint_opt to train my model, I am curious about is Constraint_opt work well with the amp, shall I make any modification?

Hi, thanks for the great work! I want to make some changes to the supernet, but I found the dynamic network is not for an original VIT. So I wonder whether there is a code for original VIT.

Hi author,
Looks only training code is released?

After skimming through the code, I cannot find the relevant part of gradient projection as Eq. 1 in the paper.

If it's my carelessness, could you please help me figure it out?

Hello,
I have been following the work related to using NAS for Vision Transformers. I am reading your approach from the NASViT paper, it would be very helpful if you could provide a documentation for your module and add descriptive comments in the code.

Hi,
Thank you for your work!
I am trying to reproduce some of the results mentioned in the paper, could you please share the NASVIT (A0-A5) checkpoints as well?

Hi, will NasVit add detection result?

I wonder if the the authors had any experiments conducted using this work on different search spaces apart from the NASViT, such as ResNet or others.

Could you please clarify this?

Thank you

Hi, when I want to use the function

compute_active_subnet_flops() 

in the model class

attentive_nas_dynamic_model.py

It will output an error said that NoneType has no active_expand_ratio. I guess the problem is this is desgined for mb and I did not find the logic of logging flops of transformer block.