Skip to content

Link the M4Bench paper on arXiv and publish Croissant 1.0.1 - #66

Merged
hannesill merged 2 commits into
mainfrom
claude/m4bench-arxiv-link
Oct 28, 2026
Merged

hannesill merged 2 commits into
mainfrom
claude/m4bench-arxiv-link

Conversation

@hannesill

Copy link
Copy Markdown
Owner

Summary

Links the published paper (arXiv:2610.37459) and brings main up to the Croissant file the Hugging Face dataset has served since 23 October.

.gitattributes                      new: Croissant JSON marked linguist-generated
CITATION.cff                        + doi 10.48550/arXiv.2610.37459, + arXiv url
README.md                           + "Paper" bullet in the M4Bench section; evidence bullet shortened
benchmark/release/v1/
├── croissant/…croissant.json       1.0.0 → 1.0.1, identical to the hosted copy
├── scripts/make_croissant.py       record sets read metadata/run-indexes/*.csv directly
└── huggingface/README.md           lists the run-index folder; links arXiv
  • 8fd20cb Croissant 1.0.1, exactly as hosted (JSON, script, dataset card).
  • 5b8c5b3 arXiv links. README.md is over the 5,000-token budget, so the evidence bullet was shortened to keep the file from growing (net −5 bytes).

Evidence

  • Before: main holds Croissant 1.0.0; Hugging Face serves 1.0.1.
    After: make_croissant.py <evidence package dir> regenerates the JSON, and cmp against huggingface.co/datasets/hannesill/m4bench-evidence/raw/main/m4bench_evidence.croissant.json reports no difference. The dataset card in 8fd20cb is also byte-identical to the hosted card.
  • cffconvert --validate: "Citation metadata are valid according to schema version 1.2.0."
  • https://doi.org/10.48550/arXiv.2610.37459 redirects to arxiv.org/abs/2610.37459; the arXiv record's title and eight authors match CITATION.cff.
  • pre-commit, including the full pytest suite, passed.

Merge Danger

Door: two-way

Docs and release metadata only. The tag m4bench-neurips2026-camera-ready and its GitHub release do not move; the tag keeps Croissant 1.0.0, which matched the dataset on release day.

Blast Radius: none

No benchmark code changes. After merge, the dataset card on main carries the arXiv link that the hosted card lacks until it is uploaded again.

The Hugging Face dataset has served this version since 23 October 2026.
Its five record sets now read the run-index CSV files from
metadata/run-indexes/ on the dataset instead of from inside the
evidence archives, and the dataset card lists that folder. The archives
and their SHA-256 checksums are unchanged.

make_croissant.py regenerates the hosted Croissant file byte for byte.
The file is marked linguist-generated so the token-limit hook skips it
and GitHub collapses its diff.
The paper is arXiv:2610.37459. The README's M4Bench section and the
dataset card link it, and CITATION.cff gives its DOI and URL, so
GitHub's "Cite this repository" points to the paper. The evidence
bullet in the README is shorter to keep the file within its size budget.
@chatgpt-codex-connector

chatgpt-codex-connector Bot commented Oct 28, 2026 •

Copy link
Copy Markdown

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review Status Commit Review trigger
📝 Code Review ✅ Completed 2026-10-28T13:54:07.318114Z 5b8c5b3 PR opened
ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

@hannesill
hannesill merged commit 72eba1e into main Oct 28, 2026
16 checks passed
@hannesill
hannesill deleted the claude/m4bench-arxiv-link branch October 28, 2026 14:37
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant