Vidushee/iclr-rejected-papers-with-code-1k

Rejected ICLR Papers with Reviews and Code

This dataset contains 1,000 rejected ICLR submissions from 2018–2026. Each row has the OpenReview submission metadata and reviews, the rejected submission PDF, and a commit-pinned archive of a matched public GitHub repository.

This collection was built directly from OpenReview. It does not use a third-party ICLR review dataset.

Project repository: TheAppliedScientist

Contents

• —1,000 unique rejected OpenReview submissions
• —1,000 unique canonical GitHub repositories
• —3,740 complete official-review records
• —1,000 PDFs downloaded from immutable OpenReview attachment paths
• —1,000 commit-pinned GitHub repository ZIP archives
• —SHA-256 checksums for every PDF and repository archive

Files

• —dataset.jsonl: full records, including all review content
• —manifest.jsonl: lightweight records without the large reviews arrays
• —report.json: final integrity and distribution report
• —papers/<forum>.pdf: OpenReview submission PDFs
• —repos/<forum>.zip: commit-pinned GitHub source archives

Paths in each row are relative to the dataset repository.

Example

from datasets import load_dataset

full = load_dataset(
    "Vidushee/iclr-rejected-papers-with-code-1k",
    "full",
)
manifest = load_dataset(
    "Vidushee/iclr-rejected-papers-with-code-1k",
    "manifest",
)

Licensing and responsible use

This repository is an aggregation of materials with mixed upstream terms. It does not relicense individual papers, reviews, or code repositories. Paper and review rights remain governed by OpenReview and the authors' terms; source-code archives retain the licenses (if any) specified by their upstream repositories; and the Papers With Code mapping archive is CC BY-SA 4.0. Users are responsible for checking the applicable upstream license before redistributing or modifying an individual artifact.

Use the OpenReview forum URL and GitHub repository URL stored in each row when citing individual works.