Inconsistency in the CIFAR10 accuracy with NFT

[SharanB7]

Hello authors,

I've tried implementing the code and got the mentioned accuracies for MNIST and FashionMNIST datasets using NFT. However, the model accuracy on the CIFAR10 dataset is 45.68 % compared to 63.4 % as stated in the paper. Passing nft in the model as you did for the other two datasets improved the accuracy to 54.67 %. Please guide me with any changes you missed while updating the code.

Thanks for your time.

Best Regards,
Sharan Bajjuri