Skip to content

Data, artifacts, and models

Use a Dataitem when the object is a typed dataset that should be understood by data-aware components (e.g. a pandas dataframe). Use an Artifact when you need to store and move a file or another binary object. Use a Model when you need to store and manage a machine learning model with model-specific metadata.

flowchart LR
    input["Data or file"] --> decision{"What are you managing?"}
    decision -->|Typed dataset| dataitem["Dataitem"]
    decision -->|File or binary object| artifact["Artifact"]
    decision -->|Machine learning model| model["Model"]
Dataitem Artifact Model
Represents A typed dataset A file or binary object A machine learning model
Choose it when The data kind and dataset semantics matter File storage and transfer are the primary concerns Model metadata and model-specific operations matter
Supported kinds table, croissant, dataitem artifact model, mlflow, sklearn, tvm-ir, tvm-so
Typical operations Load data with kind-specific methods, upload, download, represent as dataframe Upload, download, and access as a file Upload or register models, move model files, and log metrics