Artifacts
The artifact system that versions datasets, models, and intermediate outputs while tracking your pipeline.
Store datasets and models as versioned artifacts, automatically recording the lineage of which data produced which model. Content addressing that avoids duplicate storage makes even large data economical to version.
What Is Artifacts
Reproducible AI requires managing not just code but the versions of your data and models. W&B Artifacts stores datasets, models, and outputs as versioned objects and links them to experiments and models, automatically capturing the lineage of the entire pipeline.
Key Capabilities
Data & model versioning
Store datasets and models as versioned artifacts
Lineage graph
Automatically visualizes the lineage from input data through experiments to models
Efficient storage
Content addressing and deduplication make even large data economical to version
Pipeline tracking
Links outputs across preprocessing, training, and evaluation stages
What It's Used For
Data governance
Track and audit the data versions used in training
Reproducibility
Reproduce results with identical data and model versions
MLOps pipelines
Manage the lineage of data-to-model pipelines