Skip to content

Artifacts¤

The Artifacts page provides a comprehensive interface for exploring all types of artifacts (datasets, models, metrics, step metrics, labels) tracked by CMF across your ML pipelines. This page enables users to search, filter, and analyze artifacts with detailed metadata and version history.

Artifacts represent the data entities in your ML pipeline:

  • Datasets: Training data, test data, validation sets, feature matrices. Tracked with version control and DVC metadata.
  • Models: Trained ML models, model checkpoints, exported models with versioning information.
  • Metrics: Execution-level metrics (coarse-grained tracking) without backing physical files. Used to log metadata associated with the entire execution.
  • Step_Metrics: Fine-grained metrics stored in Parquet files. Captures per-step or per-epoch metrics during training/execution, committed to version control.
  • Labels: Data labels and annotations, usually CSV files containing information about datasets. Connected to datasets via "has_label" relationship.

CMF Artifacts Page

Page Features¤

1. Filter Panel¤

The filter panel allows you to narrow down artifacts based on multiple criteria:

Filter Type Description Options
Artifact Type Filter by artifact category Dataset, Label, Step_Metrics, Model, Metrics
Search Filter Search across all artifact properties Enter any text to search execution names, stages, Git commits, or any execution metadata. Matching executions will be displayed in the table.

Usage:

  1. Select a pipeline from the sidebar to view its artifacts
  2. Choose artifact type (Dataset/Label/Step_Metrics/Model/Metrics) from tabs
  3. Use filter box to search by artifact name or properties

2. Artifacts Table¤

The main table displays artifacts with the following columns:

Column Description
+ Expandable icon to view detailed artifact information
ID Unique artifact identifier
Name Artifact name and identifier
Execution Type Execution name where artifact was created
Date Timestamp of creation
URI Artifact location/path
URL Associated URL reference
Git Repo GitHub Repository URL
Commit Git commit hash

Interactions:

  • Click + icon: Expands row to show detailed artifact metadata, custom properties, and version information
  • Click column headers: Sort by NAME or DATE column (ascending/descending)
  • Pagination controls: Navigate through large artifact lists using Previous/Next buttons and page numbers

Using the Artifacts Page¤

Example 1: Find All Datasets from a Pipeline¤

  1. Navigate to Artifacts page by clicking on the header tab
  2. Select a pipeline from the LIST OF PIPELINES sidebar
  3. The Dataset tab is selected by default
  4. Review the list of all datasets used in that pipeline
  5. Click the + icon to view detailed artifact metadata

Example 2: View Models and Their Execution Context¤

  1. Select a pipeline from the sidebar
  2. Click on the Model tab to filter by model artifacts
  3. Review the EXECUTION TYPE column to see which pipeline stage created each model
  4. Click the + icon to view training parameters, version information, and metrics

Example 3: Track Metrics Over Time¤

  1. Select the Metrics or Step_Metrics artifact type tab
  2. Click the DATE column header to sort chronologically
  3. Click the + icon on any metric to view its values
  4. Compare metrics across different execution runs

Example 4: Find Artifacts with Labels¤

  1. Navigate to any artifact type tab (typically Dataset)
  2. Check the LABEL column to identify artifacts with associated labels
  3. Click the Label tab to view all label artifacts
  4. Use the filter box to search for specific label files

Additional Snapshots¤

1. Model Card¤

Model Card

2. Label Content¤

Label Content