await carton.pack(
    # ...
    model_name = "super_awesome_test_model",
)

await carton.pack(
    # ...
    short_description = "A short description that should be less than or equal to 100 characters.",
)

MODEL_DESCRIPTION = """
Some description of the model that can be as detailed as you want it to be

This can span multiple lines. It can use markdown, but shouldn't use arbitrary HTML
(which will usually be treated as text instead of HTML when this is rendered)

You can use images and links, but the paths must either be complete https
URLs (e.g. https://example.com/image.png) or must reference a file in the misc folder (e.g `@misc/file.png`).
See the `misc_files` section of this document for more details
"""

await carton.pack(
    # ...
    model_description = MODEL_DESCRIPTION,
)

await carton.pack(
    # ...
    license = "Apache-2.0",
)

await carton.pack(
    # ...
    repository = "https://github.com/VivekPanyam/carton",
)

await carton.pack(
    # ...
    homepage = "https://carton.run",
)

await carton.pack(
    # ...

    # The contents of this list are target triples. For example:
    # x86_64-unknown-linux-gnu
    # aarch64-unknown-linux-gnu
    # x86_64-apple-darwin
    # aarch64-apple-darwin
    required_platforms = [],
)

from cartonml import TensorSpec

input_x = TensorSpec(
    # The name of the input or output
    name="x",

    # The datatype of the tensor.  Valid values are
    # - float32
    # - float64
    # - string
    # - int8
    # - int16
    # - int32
    # - int64
    # - uint8
    # - uint16
    # - uint32
    # - uint64
    dtype = "float32",

    # A shape of none means the tensor can be of any shape
    shape = None,

    # A shape with 3 dims of any value
    # Note: this also supports ragged/nested tensors
    # shape = [None, None, None],

    # A shape of an exact size
    # shape = [2, 3, 512, 512],

    # A scalar
    # shape = [],

    # A shape containing a "symbol"
    # shape = ["batch_size", 3, 512, 512],

    # A symbol is an arbitrary string that should resolve to the same value each time it's used.
    # (i.e. at runtime, every dimension of size "batch_size" should
    # resolve to the same concrete value)

    # A symbol for the overall shape:
    # shape = "input_shape",

    # Optional
    description = "The input image as a NCHW tensor"
)

await carton.pack(
    # ...
    inputs = [input_x],
)

await carton.pack(
    # ...
    outputs = [],
)

from cartonml import SelfTest
import numpy as np

st = SelfTest(
    # Optional
    name = "a_self_test",

    # Optional
    description = "A self test",

    # A dict mapping input tensor names to values
    inputs = dict(x = np.ones(5, dtype=np.float32)),

    # Optional
    # If specified, the output will be compared using
    # something conceptually similar to np.allclose
    # Note: this comparison is not currently implemented, but may be in the future
    expected_out = dict(out = np.ones(5, dtype=np.float32) * 2)
)

await carton.pack(
    # ...
    self_tests = [st],
)

from cartonml import Example
import numpy as np

# Read an input image
with open('/path/to/input.png', 'rb') as f:
    input_image = f.read()

e = Example(
    # Optional
    name = "an_optional_name",

    # Optional
    description = "An example for classification of an image",

    # A dict mapping input tensor names to values
    inputs = dict(x = input_image),

    # A dict mapping output tensor names to values
    sample_out = dict(out = np.array(5))
)

await carton.pack(
    # ...
    examples = [e],
)

# ...

MODEL_DESCRIPTION = """
This model has an interesting architecture:

![Model Architecture](@misc/model_architecture.png)
"""

async def main():
    with open('/path/to/model_architecture.png', 'rb') as f:
        model_architecture = f.read()

    await carton.pack(
        # ...
        model_description = MODEL_DESCRIPTION,
        misc_files = {
            "model_architecture.png": model_architecture
        }
    )