WRSWorld Research StateContribute
WORLD RESEARCH STATE

DataComp: fixed-recipe multimodal dataset experiments

Contribution grade

Multimodal dataset evaluation · Human-designed research method or resource

Reference score115,000Provisional editorial estimate of this exact contribution; its grade is assessed separately.

What changed

Provided candidate image-text pools, fixed CLIP training/evaluation recipes and compute-scaled tracks, with DataComp-1B demonstrating the importance of filtering under controlled training conditions.

Scientific scope

Provided candidate image-text pools, fixed CLIP training/evaluation recipes and compute-scaled tracks, with DataComp-1B demonstrating the importance of filtering under controlled training conditions. Inherits CLIP and web-pair collection rather than inventing them. Ninefold compute comparison involves a larger LAION-trained architecture; not a universally hardware-matched data-only speedup. Fixed test suites do not guarantee uncontaminated or universal real-world generalization.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Inherits CLIP and web-pair collection rather than inventing them. Ninefold compute comparison involves a larger LAION-trained architecture; not a universally hardware-matched data-only speedup. Fixed test suites do not guarantee uncontaminated or universal real-world generalization.

Contributors

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction