Models

Models is a feature of the Code Ocean platform designed to give computational scientists and ML practitioners a unified environment for developing, training, tracking, and registering machine learning models. With native MLflow integration built directly into the platform, teams working in life sciences, biotech, and pharma can manage the full ML lifecycle — from experimentation through to pre-production validation — while maintaining a traceable, reproducible record of all data and model provenance.

Models is purpose-built for AI, Machine Learning, Deep Learning, and Generative AI workflows in computational science. It integrates seamlessly with the broader Code Ocean platform, including Compute Capsules, Pipelines, the Lineage Graph, and no-code Apps, enabling both coding and non-coding users to participate in model development and validation without friction.

Key Capabilities

• Native MLflow integration: MLflow is natively embedded into the Code Ocean platform, requiring no additional setup. Users can click to track model development, parameters, and experiments directly from any Capsule, managing models from development through to production with out-of-the-box reproducibility.
• Hugging Face model import: A built-in Hugging Face integration allows users to import HF models and their associated metadata directly into a Code Ocean asset. Imported models can be used in Capsules or Pipelines and automatically benefit from provenance tracking and lineage recording.
• Flexible GPU and compute management: Users can define specific CPU, GPU, and RAM requirements for every step of the model development process. GPU-ready environments can be installed and provisioned in a few clicks, and cloud resources can be scaled up or down as needed using reliable, Dockerized environments.
• Full data and model lineage: The Lineage Graph feature enables users to trace exactly what dataset a registered model was trained on and what code was executed during development — all retrievable in seconds through an automated, immutable provenance record.
• No-code inference apps for non-coders: Models can be released as no-code inference applications with a single click, enabling domain experts and non-coding users to run inference in the cloud independently. Users can upload their own data and run the model without any coding knowledge, supporting confident pre-production validation before pushing to production.

How Models Fits Into the Code Ocean Platform

• Develop in Compute Capsules: Compute Capsules provide a flexible, reproducible environment for building model environments, assigning compute resources, and tracking experiments via MLflow. Each Capsule encapsulates code, data, environment, and generated results.
• Validate with no-code Apps: Once a model reaches a pre-production state, it can be packaged into a no-code App so that domain experts can test it on real data by uploading files and clicking Run — without requiring developer support.
• Trace provenance with the Lineage Graph: The Lineage Graph provides a visual representation of every Capsule, Pipeline, and Data asset involved in a computation, giving teams a complete and automated audit trail of model and data provenance.

Security and Pre-Production Workflow

• Secure MLflow integration: Because MLflow is natively integrated into Code Ocean rather than added as an external tool, it inherits the platform's existing security features, including industry-standard permissions and access management with identity provision and single sign-on (IDP and SSO).
• Pre-production validation: Code Ocean addresses the pre-production gap — the bridge between development and production where domain experts test a finished model on real data — by packaging the model, its environment, and a simple UI into a shareable asset. Domain experts can upload data, click Run, and evaluate results with minimal steps.

Models is part of the broader Code Ocean platform, which also supports data analysis, bioinformatics pipelines, multiomics, imaging, and cloud management. The platform runs on AWS Batch out-of-the-box, supports spot instances and automated shutdown for cost management, and is built to meet FAIR data principles. It is suitable for computational scientists, R&D leadership, IT and engineering teams, and bench scientists across life sciences organisations.