UmJCvpsFmyy5Jbro56q4JEosWCAi9uXVeW3SSBW4jbu3plJNMrnBvBgPM8TI8G3+
5hEGf2vHkgXOGMk=
=nYS8
-----END PGP PUBLIC KEY BLOCK-----

pub    37E9D58EA4598641
uid    Dain Sundstrom <dain@iq80.com>

sub    38185785755267BD
sub    381E922B6E350BF1
-----BEGIN PGP PUBLIC KEY BLOCK-----
Version: BCPG v1.68

mQENBE4KeqwBCADHUgytFpv7cJv8wfVQwyuYC8+QtbdTtOErEIpnaYQT9FV9uAs8

it downstream (when targeting a different environment, like TensorFlow lite for
example).
The MLIR module is expected to have a main() function that will be used as an
entry point. The inputs to the operations will be passed as argument to the
main() function and the returned values of the main function mapped to the
outputs.
Example usage:

```
import tensorflow as tf

    of other integer types) in `tf.constant`.
*   Make the `gain` argument of convolutional orthogonal initializers
    (`convolutional_delta_orthogonal`, `convolutional_orthogonal_1D`,
    `convolutional_orthogonal_2D`, `convolutional_orthogonal_3D`) have
    consistent behavior with the `tf.initializers.orthogonal` initializer, i.e.
    scale the output l2-norm by `gain` and NOT by `sqrt(gain)`. (Note that these

            "type": "string"
          },

							Reason: "PodInitializing",
						},
					},
				}},
				ContainerStatuses: []core.ContainerStatus{{
					ContainerID: "docker://numbers",
					Image:       "nginx:alpine",
					Name:        "main",
					Ready:       false,
					Started:     proto.Bool(false),
					State: core.ContainerState{
						Waiting: &core.ContainerStateWaiting{
							Reason: "PodInitializing",
						},
					},