---
pretty_name: CausalMix v1
configs:
- config_name: small
  data_files:
  - split: test
    path: "graphs/small/*.csv"
- config_name: large
  data_files:
  - split: test
    path: "graphs/large/*.csv"
- config_name: synthetic
  data_files:
  - split: test
    path: "graphs/synthetic/*.csv"
---

# Anonymous Data

This dataset contains benchmark files for evaluating language models on causal-discovery tasks.

The `small` and `synthetic` configs are available through the Hugging Face dataset viewer. The `large` config is provided as direct CSV files under `graphs/large/`; these files are large enough that automatic viewer/Croissant conversion may omit them, so the NeurIPS Croissant metadata is completed manually from the hosted CSV file layout.