Constarium
← Search

Table · dataset · 2026

CapBencher

Listed in Hugging Face Datasets

Description

CapBencher: Give Your LLM Benchmark a Built-in Alarm for Test-Set Overfitting ICML 2026 | arXiv:2505.18102 | Code | Blog Post CapBencher is a simple protocol for "capping" an LLM benchmark's accuracy by design. It sets a ceiling on the best achievable score, so that statistically significant performance above that cap becomes a strong signal of data leakage, contamination, or leaderboard hacking. A benefit is that it enables open, reproducible evaluation and model ranking… See the full description on the dataset page: huggingface.co/datasets/ishidalab/capbencher.

Links

Where it is published

Documentation and papers

Catalogue records · 1

Topics

Provenance · 1 source records, 11 field assertions
SourceKeyLast seenRaw
Hugging Face Datasetsishidalab/capbencher12 d agoJSON v1
FieldAssertionExtractorEvidence
access_levelsource · Hugging Faceconnector:huggingface@1.0.0/gated
concepts[field].local:field:computer-science-aimapping · Hugging Faceconnector:huggingface@1.0.0
concepts[modality].hf_modality:textsource · Hugging Faceconnector:huggingface@1.0.0
concepts[task].hf_task:question-answeringsource · Hugging Faceconnector:huggingface@1.0.0/tags[task_categories:*]
concepts[task].hf_task:text-generationsource · Hugging Faceconnector:huggingface@1.0.0/tags[task_categories:*]
created_datesource · Hugging Faceconnector:huggingface@1.0.0
descriptionsource · Hugging Faceconnector:huggingface@1.0.0/description
licensesource · Hugging Faceconnector:huggingface@1.0.0/tags[license:*]
publication_datesource · Hugging Faceconnector:huggingface@1.0.0
titlesource · Hugging Faceconnector:huggingface@1.0.0/id
updated_datesource · Hugging Faceconnector:huggingface@1.0.0