The cnvrg.io platform enables the following elements of ML deployment and management:
- Creation of ML pipelines—Pipelines are key to the ML processes as raw data is ingested, cleansed, and transformed for use in modeling. cnvrg.io provides a drag-and-drop graphical interface for constructing and managing end-to-end AI and ML pipelines. These pipelines span from training to production and include dashboards for monitoring the state of jobs and resources, code reusability, and traceability.
- AI library management—The cnvrg.io AI Library is a native package manager for model code that promotes both code reusability and collaboration providing all the necessary container management for software from various open-source and OEM-provided registries. Data scientists can choose from open-source models or their own models through GitHub and Bitbucket integration. Changes can be pushed back to the remote repository when experimentation is complete.
- Heterogeneous compute—cnvrg.io abstracts compute and storage hardware into a cloud-like utility, making it easy to deploy jobs to various compute and storage resources with seamless management across environments. This capability means that locality of data or even locality of ML processing is not important, giving researchers more flexibility to handle and work with data or ML models in a location-agnostic manner.
- Centralized dataset control and centralized version control—Enables data scientists to connect and share any data from any source, creating datasets in a storage-agnostic manner. This centralized version-controlled system tracks every stage and all actions are committed, benefiting version control, explainability, and regulatory compliance.
- Orchestrator and workload scheduler—Uses any Kubernetes distribution such as VMware Tanzu, Symworld Cloud Native Platform, RedHat OpenShift, SUSE Rancher, or one that is compliant with Cloud Native Computing Foundation (CNCF) as an orchestration, scheduling, and scaling layer. It makes jobs portable across environments and scales pods and clusters up and down on demand. cnvrg.io also uses Kubernetes’ own native mechanisms, such as taints and tolerations, to place workloads only on appropriate nodes.
- Machine learning tracking—Automatically tracks and stores model code, statistics, and artifacts, bringing better control over the process with easy reproducibility and comprehensive monitoring. Real-time monitoring and interaction enable greater accuracy, performance, and reproducibility of models.
- Machine learning model deployment—Reliable deployment with automatic monitoring for zero down time. Deploy any ML model with a single click using various deployment models or interfaces.
- Scalable inferencing endpoints—Deploy large-scale and real-time machine learning to production in one click. cnvrg.io enables developers to serve models in batches, with RESTful APIs, or with streaming endpoints for real-time use cases.