dgl.to_half¶

dgl.to_half(g)[source]¶

Cast this graph to use float16 (half-precision) for any floating-point edge and node feature data.

A shallow copy is returned so that the original graph is not modified. Feature tensors that are not floating-point will not be modified.