#numerics

PyTorch Inductor returns NaNs on zero-size interpolate

Compiled F.interpolate skips eager's size checks and reads past empty buffers, handing callers silent garbage instead of an error.