{d: 5 * 365 * 24 * time.Hour, want: "5y"},
		{d: 6 * 365 * 24 * time.Hour, want: "6y"},
		{d: 7 * 365 * 24 * time.Hour, want: "7y"},
		{d: 8*365*24*time.Hour - time.Millisecond, want: "7y364d"},
		{d: 8 * 365 * 24 * time.Hour, want: "8y"},
		{d: 8*365*24*time.Hour + 364*24*time.Hour, want: "8y"},
		{d: 9 * 365 * 24 * time.Hour, want: "9y"},
	}
	for _, tt := range tests {

    based on the cost. As the size of the kernel is relatively small, this won't
    add additional cost to TPU device time. Below is the logic to transform the
    kernel of shape [7, 7, 3, 64] to [4, 4, 12, 84].

    ```python
    conv0 = tf.compat.v1.layers.Conv2D(
     filters=filters,
     kernel_size=kernel_size,
     strides=2,
     padding=('SAME' if strides == 1 else 'VALID'),