[FEA] Add example of workflow deployment using metaflow

[raybellwaves]

Is your feature request related to a problem? Please describe.
Could add an example of a workflow deployment using metaflow at https://docs.rapids.ai/deployment/stable/examples/

Describe the solution you'd like
It could be a notebook but metaflow works with .py files. The notebook could still contain the code to generate the .py file and how to run it.

Describe alternatives you've considered
NA

Additional context
I've got experience with this. Happy to create a PR. Folks at outerbounds may have thoughts as well cc. @hugobowne @tuulos @emattia or the notebook simple cross-post to https://outerbounds.com/blog/nvidia-cloud-gpu-announcement/ or other documents on using GPU's with metaflow

[emattia]

Nice idea @raybellwaves, lmk how I can help. Here is a starter template I've used recently:

from metaflow import FlowSpec, step, kubernetes
from metaflow.profilers import gpu_profile


class RapidsFlow(FlowSpec):

    @step
    def start(self):
        self.next(self.train)

    @gpu_profile(interval=1)
    @kubernetes(
        cpu=4,
        memory=22000,
        gpu=1,
        image="docker.io/rapidsai/rapidsai:23.06-cuda11.8-runtime-ubuntu22.04-py3.10",
    )
    @step
    def train(self):
        """https://docs.rapids.ai/api/cuml/stable/estimator_intro/"""

        import cuml
        from cuml.datasets.classification import make_classification
        from cuml.model_selection import train_test_split
        from cuml.ensemble import RandomForestClassifier as cuRF

        n_samples = 5_000_000
        n_features = 100
        n_classes = 2

        # random forest depth and size
        n_estimators = 25
        max_depth = 10

        # generate synthetic data [ binary classification task ]
        X, y = make_classification(
            n_classes=n_classes,
            n_features=n_features,
            n_samples=n_samples,
            random_state=0,
        )

        self.X_train, self.X_test, self.y_train, self.y_test = train_test_split(
            X, y, random_state=0
        )

        model = cuRF(max_depth=max_depth, n_estimators=n_estimators, random_state=0)

        self.trained_RF = model.fit(self.X_train, self.y_train)
        self.predictions = model.predict(self.X_test)

        self.next(self.eval)

    @kubernetes(
        cpu=1,
        memory=12000,
        gpu=1,
        image="docker.io/rapidsai/rapidsai:23.06-cuda11.8-runtime-ubuntu22.04-py3.10",
    )
    @step
    def eval(self):
        import cuml
        from cupy import asnumpy
        from sklearn.metrics import accuracy_score

        cu_score = cuml.metrics.accuracy_score(self.y_test, self.predictions)
        sk_score = accuracy_score(asnumpy(self.y_test), asnumpy(self.predictions))

        print(" cuml accuracy: ", cu_score)
        print(" sklearn accuracy : ", sk_score)

        self.next(self.end)

    @step
    def end(self):
        pass


if __name__ == "__main__":
    RapidsFlow()

Could cross-post an example on https://outerbounds.com/docs/how-to-index/, or write a blog if we do something more of a real-world reference.