DEVOPSdigest

Dotscience announced its designation as a GitLab Technology Partner.

As a GitLab Technology Partner, Dotscience is extending the use of its platform for collaborative, end-to-end ML data and model management to the more than 100,000 organizations and developers actively using GitLab as their DevOps platform.

Available as SaaS or on-prem, Dotscience dramatically simplifies the deployment and monitoring of ML models, making MLOps accessible to every data scientist without the need to set up and configure complex and powerful tools like Kubernetes, Prometheus and Grafana from scratch. GitLab users can now easily and quickly deploy Dotscience directly through the GitLab Technology Partner application library.

Brandon Jung, VP of Alliances at GitLab, said: “GitLab users who are looking to make the deployment of machine learning models as easy, fast and safe as software engineering is with DevOps should explore Dotscience in combination with GitLab to gain the most value out of their AI initiatives.”

ML is the next wave of computing, and in order for it to be successful, DevOps best practices need to be applied to ML development. Dotscience’s integration with GitLab enables data science and ML teams to customize and hand off the model build stage to a GitLab pipeline, enabling the creation of more powerful MLOps pipelines.

The GitLab Technology Partner program features a wide selection of business applications from a variety of partners, including cloud and Kubernetes technologies. GitLab’s specialized alliances team also helps customers compare options and quickly find applications to fit their business needs.

“More businesses are increasingly looking for solutions and tools to improve ways to operationalize AI and measure the business value of their AI initiatives,” said Luke Marsden, CEO and founder at Dotscience. “For example, monitoring models in ML-specific ways is not obvious to software-focused DevOps teams, neither is the need to track not only code versions but also data versions, model versions and the provenance graph of relationships between code, data and models, which we call ‘runs’. The Dotscience platform tracks runs and offers ML-specific statistical monitoring, enabling ML and data science teams to achieve reproducibility, accountability, collaboration and continuous delivery across the AI model lifecycle.”

Dotscience manages data versioning, model training, model versioning, deployment and monitoring. As part of the model deployment stage, it's necessary to convert versioned model files (e.g., a Tensorflow neural network and its weights, serialized as protobufs) into a runnable microservice which takes requests (e.g., pictures of road signs) and returns a classification (e.g., "this is a stop sign").

Dotscience comes with a built-in model builder, which comes with some hard-coded Dockerfiles. What if you want to customize that Docker image, add some extra libraries, data pre-processing or make it work in a different way? For example, you might want to make the model operate on inputs from a queue, rather than listening on a REST endpoint. Or you might want to connect multiple models together. Or you might want to automatically run tests against the model before pushing it to the container registry.

The Dotscience GitLab integration enables all of these use cases. By passing off the model build stage to a GitLab pipeline, teams can customize this process and build more powerful MLOps pipelines.