Haifan Gong requests CTGroundBench benchmark addition to Hugging Face Hub

Haifan Gong has submitted a request to the Hugging Face team to add the haifan-gong/CTGroundBench dataset to the official Benchmark allow-list. This addition would enable the aggregation of community evaluation results for language models on clinical CT data.

• CT-GroundBench evaluates grounding of free-text CT findings in phenotype relationships, attenuation intervals, and segmentation counts.
• It defines three native leaderboard tasks: task1_accuracy, task2_interval_iou, and task2_region_accuracy.
• The dataset uses the inspect-ai evaluation framework and passed Hub push-time validation.
• It includes ground truth, 14 prediction files from seven models, and SHA-256 checksums for auditability.

The author seeks to establish an official leaderboard for clinical phenotype grounding evaluations.