pytorch

45079d94 - Output allocate patch (#1790)

Output allocate patch (#1790) Caching strides along with sizes. This is to support current expand, which introduces non-contiguous output tensor