Is it possible for Keras to load tensors directly into VRAM and skip RAM?

Question

Suppose I have a machine with very little RAM but it has a beefy GPU, is it possible to bypass RAM and load a .npy file directly into GPU?

For example,

<