datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
fda-facility-compliance-intelligence
FDA Facility Compliance Intelligence
Version: 1.0.0 | Records: 132,080 | Price: $3,500 | Source: FDA (public domain)
Dataset Summary
The dataset's core signal — whether a facility's inspections escalated to a Warning Letter — tracks FDA's own severity classifications: facilities whose most recent inspection was OAI have escalated to a Warning Letter 65.7% of the time, versus 2.5% for NAI — a ~26× relationship you can reproduce directly from this file (GROUP BY… See the full description on the dataset page: https://huggingface.co/datasets/RubyIntelligence/fda-facility-compliance-intelligence.CoT-ComplianceBench
CoT-ComplianceBench
CoT-ComplianceBench is the first benchmark that directly audits the chain-of-thought (CoT) reasoning traces of large language models under adversarial prompting — independently of their final outputs.
We introduce the Compliance Divergence Rate (CDR): the fraction of adversarial prompts where a model reasons unsafely in its CoT yet produces a safe-looking final response, creating a blind spot undetectable by output-only evaluators. We also introduce the… See the full description on the dataset page: https://huggingface.co/datasets/BharathGaddam/CoT-ComplianceBench.aion-cr-compliance-dataset