Request access to the ALE task data archive

This dataset bundles each task's input, software, and reference (ground-truth) data into a single archive for convenient one-shot download — in particular for running ALE locally with the local Docker provider. Because it includes the reference outputs used to score runs, access is gated to protect benchmark integrity. Approval is granted manually.

By requesting access you agree to ALL of the following. Misuse may result in access revocation and exclusion from the benchmark leaderboard.

Log in or Sign Up to review the conditions and access this dataset content.

Agents Last Exam — Task Data Archive (input + reference)

⚠️ Gated dataset. This repo packages each task's input, software, and reference (ground-truth) data into a single archive (ale-tasks-data.tar.gz) for convenient one-shot download — in particular for running ALE locally with the local Docker provider, which fetches it and mounts each task's data at run time. Because it includes the reference outputs used to score runs, access requires login, agreement to the terms on the access-request form, and manual approval.

The ALE dataset family

Dataset Contents Access
Task card metadata One row per task: titles, prompts, taxonomy Open
Task input data The input/ files each task hands the agent Open
Reference (ground-truth) data The reference/ outputs used for scoring ⚠️ Gated
Task data archive (this repo) input/ + software/ + reference/, bundled in one archive ⚠️ Gated

When to use this repo

Use this archive if you want everything for every task in one download — especially to run ALE locally with the local Docker provider. If you only need the open input files, use the input data repo; the reference repo holds the ground-truth alone.

Download (after approval)

huggingface-cli login
huggingface-cli download agents-last-exam/agents-last-exam-data-archive \
    ale-tasks-data.tar.gz --repo-type dataset --local-dir .
tar xzf ale-tasks-data.tar.gz

For the local Docker provider, the repo's scripts/fetch_task_data.sh does the download + extract for you — see the guide linked above.

Layout

The archive expands to one directory per task variant:

<domain>/<task>/<variant>/
├── input/       # agent-visible materials
├── software/    # task software fixtures (when present)
└── reference/   # ground-truth outputs  (why access is gated)

Terms of Use (summary)

  • No training. Do not use this data to train, fine-tune, or distill models.
  • No leakage at eval time. Do not expose this data to any model/agent while it is being evaluated on ALE.
  • No redistribution. Do not repost or share access.
  • Access is logged (HF username, email, timestamp, submitted fields) and may be revoked on misuse.

The full terms are presented on the access-request form.

Downloads last month
27