When I run the last cell(Vision Transformer AugReg) , the ValueError happened : Parsing builder name string ./flower_photos failed. How should I set the format? Maybe I need one detail method, thank you very much! The dataset dir in the red rectangle as follows:

I'm trying to install this repo onto my Jupyter notebook but it's always this error that I get. I tried uninstalling the packages listed in vit_jax/requirements.txt and installing the repo again but that hasn't worked.

Will this Network help many-to-many sequence preditction?We've tried LSTM,BRNN,Attention layer.But those do not do well in the abrupt part of the data set.Will this work well?

I'm using jaxlib with cudnn to run the LiT Model. The inference time for a single image is around 5 seconds on Quattro 8000 48GB. I'm also new to jax so maybe I'm missing something there. On CPU (I just removed jaxlib cudnn to test this) the times are much faster to be around 2 seconds. I found that...

I was trying to go through colab notebook and found this. I think there is supposed to be momentum_clip.py file however it's missing. I was thinking accidentally i removed but it is also not in the git repo. is it obliterated or new changes brought somewhere

Hi， How can I get the accuracy after training? The program outputs the following files.

I am getting an import issue while running vit_jax.ipynb on Colab which was working fine 2 days ago. Monitoring from jax is not being imported in checkpoint.py and train.py. The jax version I'm using is 0.3.25. Please let me know regarding how to fix this,

Hello and thank you for your work on ViT! I was writing a blogpost on the ViT architecture and came across the discrepancy with standard Transformer encoder. As stated in the paper, ViT was developed with the fewest possible modifications in Transformer encoder, but I compared Transformer encoder bl...