Mamba training not enhancing performances.

Hello there,
Just wanted to ask if someone tried to train mamba from scratch. I am training it using my own dataset which is inherently imbalanced. I trained a ResNet on it and got an accuracy of 0.77, wanted to test mamba but it seems like not improving at all, i know 3 batches is a small number but getting an accuracy, recall and precision of zero for the minor class isn't weird ?

Dataset : 276k training examples
Batch …

accuracy dataset deeplearning hello mamba performances resnet test train training