Eval Cards Backend Dataset Pre computed evaluation data powering the Eval Cards frontend. Generated by the eval cards backend pipeline. Last generated: 2026 05 05T11:30:42.961096Z Quick Stats Stat Value Models 5,678 Evaluations (benchmarks) 798 Metric level evaluations 1321 Source configs processed 52 Benchmark metadata cards 240 File Structure The legacy per detail JSON directories ( models/ , evals/ , developers/ ) and their list view aggregators ( model cards .json , eval list .json , developers.json ) are no longer published — the same data lives in duckdb/v1/ .parquet under the same row shape. How to Fetch Data Base URL All files are accessible via the HuggingFace dataset file API: Access Patterns 1. Bootstrap — load the manifest and metadata JSONs 2. Query model / eval / developer data via DuckDB Each parquet row carries a payload json column holding the same shape that the legacy model cards.json / eval list.json / models/{id}.json / evals/{id}.json / developers/{slug}.json files used to expose. Read with DuckDB or any Parquet client. 3. Get benchmark metadata card Lookup key: use the benchmark leaf key from any eval summary. 4. Get peer rankings Rankings are keyed by eval s…
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy