Inconsistency in handling of DT_FLOAT and DT_DOUBLE types by gradient calculation

src/TensorFlowNET.Core/Gradients/gradients_impl.py.cs

@@ -368,7 +368,7 @@ namespace Tensorflow
                     if (y.dtype.is_complex())
                         throw new TypeAccessException($"Gradients of complex tensors must set grad_ys (y.dtype = {y.dtype})");
                     var shape = array_ops.shape(y);
-                    var constant = constant_op.constant(1.0f, name: $"grad_ys_{i}");
+                    var constant = constant_op.constant(y.dtype == TF_DataType.TF_DOUBLE ? (object)1.0 : (object)1.0f, name: $"grad_ys_{i}");
                     var fill = gen_array_ops.fill(shape, constant);
                     new_grad_ys.Add(fill);
                 }