jrc-ai/fit-check-data

FIT-CHECK — Pre-built data

Pre-built, uncompressed data for the FIT-CHECK Streamlit application (Fitness & Impact Tool for Checking EU Legislation).

The FIT-CHECK app downloads these files at startup instead of baking them into the Docker image.

Contents

Path	Description	Approx. size
data/acquis.duckdb	DuckDB metadata database (acts, references, full text for BM25)	~1.15 GB
data/lancedb/	LanceDB vector store (act_chunks.lance, multilingual-e5-large embeddings)	~1.4 GB

The layout mirrors the application's local data/ directory exactly, so the app can snapshot_download(allow_patterns=["data/**"]) straight into its working directory.

Usage

from huggingface_hub import snapshot_download

snapshot_download(
    repo_id="jrc-ai/fit-check-data",
    repo_type="dataset",
    local_dir=".",            # creates ./data/acquis.duckdb and ./data/lancedb/
    allow_patterns=["data/**"],
    token="hf_...",           # required while the dataset is private
)

Models

Embeddings were produced with multilingual-e5-large via the GENESIS (GPT@JRC) API.

Contact

• Technical / FIT-CHECK: sebastien.campion@ec.europa.eu
• GENESIS API: JRC-GENESIS@ec.europa.eu

Developed by the Joint Research Centre of the European Commission.