produce the generated Tensor. For instance, if size is (4, 8)
                and steps is (1, 4), then a tensor `t` of size (4, 32) will be
                created and then `t[:, ::4]` will be used. (Allowing one to test
                Tensors with strided memory.)
            probability_contiguous:
                A number between zero and one representing the chance that the
                generated Tensor has a contiguous memory layout. This is achieved by
                randomly permuting the shape of a Tensor, calling `.contiguous()`,
                and then permuting back. This is applied before `steps`, which can
                also cause a Tensor to be non-contiguous.
            min_elements:
                The minimum number of parameters that this Tensor must have for a
                set of parameters to be valid. (Otherwise they are resampled.)
            max_elemnts:
                Like `min_elements`, but setting an upper bound.
            max_allocation_bytes:
                Like `max_elements`, but for the size of Tensor that must be
                allocated prior to slicing for `steps` (if applicable). For
                example, a FloatTensor with size (1024, 1024) and steps (4, 4)
                would have 1M elements, but would require a 64 MB allocation.
            dim_parameter:
                The length of `size` and `steps` will be truncated to this value.
                This allows Tensors of varying dimensions to be generated by the
                Fuzzer.
            dtype:
                The PyTorch dtype of the generated Tensor.
            cuda:
                Whether to place the Tensor on a GPU.
            tensor_constructor:
                Callable which will be used instead of the default Tensor
                construction method. This allows the author to enforce properties
                of the Tensor (e.g. it can only have certain values). The dtype and
                concrete shape of the Tensor to be created will be passed, and
                concrete values of all parameters will be passed as kwargs. Note
                that transformations to the result (permuting, slicing) will be
                performed by the Fuzzer; the tensor_constructor is only responsible
                for creating an appropriately sized Tensor.
        N)