Deep learning: why my network do not overfit for one simple data?

Question

I am building a neural network for my project. After 1000 epoch, the loss is very high, and the prediction is much differ from the target.

Then, I use one simple data