YaxinLuo/PosterBench-mini

PosterBench-mini

PosterBench-mini is the fixed 10-paper development subset of PosterBench, introduced in AutoDesign: Meta-Harness Optimization for Long-Horizon Agentic Design. It contains exactly two papers from each of the benchmark's five disciplines.

This Hugging Face release is metadata only. It does not host or redistribute the underlying paper PDFs, full text, abstracts, figures, tables, or thumbnails.

Load

from datasets import load_dataset

dataset = load_dataset("YaxinLuo/PosterBench-mini", split="test")

For the complete 100-paper benchmark, use `YaxinLuo/PosterBench`.

Dataset structure

The columns follow the full PosterBench manifest:

• —id, discipline, case_id
• —title, authors, year
• —identifiers
• —landing_url, download_policy
• —pdf_url, license_url
• —expected_sha256, schema_version

pdf_url and license_url are present only when redistribution rights for the exact paper version were verified. Records marked manual_access_required must be obtained independently through a lawful source available to the user.

License and rights boundary

The MIT license for this repository applies only to the PosterBench metadata, benchmark-specific annotations, and documentation authored by the AutoDesign project. It does not apply to or grant any rights in the underlying papers or their contents. Each paper remains subject to its own copyright and license. Users are responsible for confirming that their access and use are lawful.

Source and reproducibility

• —AutoDesign repository
• —PosterBench evaluation guide
• —Canonical full manifest
• —Canonical mini IDs
• —Paper page

Please cite the AutoDesign paper when using PosterBench-mini.