Skip to content

Conclusion

Congratulations! You did it! You were able to convert a ML experiment with a traditional approach to a well-defined, well-documented workflow that can scale and serve a model to the outside world! You also learned how to improve the performance of your model with additional and high-quality data in an iterative manner.

Let's take the time to make a summary of what you have done.

Summary of what you have done

  • The codebase can be shared among the developers

Thanks to Git, the codebase can be shared and improved collectively among the developers.

  • The dataset can be shared among the developers

Thanks to DVC, the dataset can be shared and improved collectively among the developers.

  • The model can be reproduced

Thanks to DVC, the steps to create the model are documented and can be executed in order to reproduce the model.

  • The experiment can be executed on a clean machine

Thanks to the CI/CD pipeline, the experiment can be executed on a clean machine. Erasing the "but it works on my machine" issue.

  • The changes done to a model can be tracked

Thanks to DVC and CML, the changes done to a model can be tracked, discussed and visualized before merging them.

  • The model can be used outside of the experiment context

Thanks to BentoML, the model can be served and be used outside of the experiment context.

  • The model can be deployed and accessed on Kubernetes

Thanks to BentoML, the model can be deployed and be accessed on a Kubernetes server.

  • The model can be trained on a Kubernetes pod

Thanks to a self-hosted runner, the model can be trained on specialized hardware on a Kubernetes pod.

  • The model performance can be improved by retraining with additional data.

Thanks to Label Studio, additional and high-quality training data can be used to retrain the model.

End of your journey

We appreciate your continued support! We trust that you found this guide both enjoyable and informative. We encourage you to explore the left sidebar for additional MLOps-related resources that we believe might further enhance your understanding and skills in this rapidly evolving and increasingly important field.

If you encounter any difficulties, please don't hesitate to reach out to us on GitHub.

Additionally, if you found our MLOps guide valuable, we would be grateful if you could take a moment to star our GitHub repository. Your support helps us improve and expand our offerings for the community. Thank you!

Happy learning! :)