When we use the C APIs to create kernels, we can create a new TF_Tensor or get from context. For the ops using generic data type such as float or int8. We can get the raw buffer pointer and do what we want.
But when we get the tensor whose type is TF_VARIANT, is there any solution that we can get the data in it and do computing ? Based on currently C APIs, we can only get some meta data of the Tensor, such as shape info, data type and so on.