I've been learning about the Transformers model and its different variations (like the Informer model, etc.) and I found something weird!
It looks like there is a level of randomness in their results. The results are different when I train the model and then test it. It might not be that significant when you check measures like MSE or MAE but it can have huge effects on choices that are made according to the model's predictions.
Why does this happen in Transformer-like models?