LoDTensorDesc::dims still use `int32`
Created by: JiayiFeng
current LoDTensor
:
message LoDTensorDesc {
required DataType data_type = 1;
repeated int32 dims = 2; // [UNK, 640, 480] is saved as [-1, 640, 480]
optional int32 lod_level = 3 [ default = 0 ];
}
Our Tensor::dims_
is using int64, so here we shall also use int64.