Artifact Bench

A heterogeneous graph of 14,053 HuggingFace model, dataset, paper, and codebase nodes. It contains 51,337 observed evaluation edges linking models to datasets with performance metrics, intended for benchmarking link prediction and attribute regression tasks. The dataset was created by user lwaekfjlk and was last updated on 2026-05-16.

Use Cases

• Benchmarking link prediction algorithms based on the heterogeneous graph structure of models, datasets, and papers.
• Training regression models to predict performance metric values based on the observed evaluation edges.
• Analyzing relationships and dependencies between different types of machine learning artifacts (models, datasets, papers).
• Validating graph-based machine learning methods using the provided agent-based verification suite.

Strengths

• Contains 14,053 nodes representing concrete machine learning artifacts.
• Includes 51,337 observed evaluation edges, providing a substantial relational structure.
• Released under the Open Database License (ODbL) v1.0, permitting sharing and modification with attribution.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for specific modeling tasks.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

HuggingFace platform, aggregated from multiple sources.

Collection Method

Likely constructed by scraping or aggregating metadata and evaluation results from the HuggingFace ecosystem.

Freshness

Last updated 2026-05-16 09:04:16; freshness should be verified.