I'm not familiar with PyTorch so this is purely a guess.

In the examples, all of the array elements are floating point numbers: that's probably what homogeneous means in this context.

By contrast, arrays with a mix of floats, integers, strings, objects, and so on, would be heterogeneous.