datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
LRM-Safety-evaluation-parsedDiNa-LRM-SD35m-HPSv3-Preprocess-DataLRMovieNetThis is the LRMovieNet dataset proposed by ECCV 2024 Paper "Multimodal Label Relevance Ranking via Reinforcement Learning". The code is available at https://github.com/ChazzyGordon/LR2PPO.
Please go to Files and versions to download the LRMovieNet dataset.
We select 3,206 clips from 219 videos in the MovieNet dataset.
For each movie clip, we extract frames from the video and input them into the RAM model to obtain image labels.
Concurrently, we input the descriptions of each movie clip into… See the full description on the dataset page: https://huggingface.co/datasets/ChazzyGordon/LRMovieNet.lrm-safety-eval
Chain of Risk — LRM Safety Evaluation Dataset
⚠️ Content Warning: This dataset contains potentially harmful, unsafe, or unethical prompts
and model responses collected strictly for safety research purposes.
Dataset Summary
This dataset accompanies the paper:
Chain of Risk: Safety Failures in Large Reasoning Models and Mitigation via Adaptive Multi-Principle Steering
Xiaomin Li, Jianheng Hou, Zheyuan Deng, Zhiwei Zhang, Taoran Li, Binghang Lu, Bing Hu, Yunhan Zhao… See the full description on the dataset page: https://huggingface.co/datasets/HJH2CMD/lrm-safety-eval.LRM-Safety-Study
LRM-Safety-Study Dataset
Dataset Description
This dataset was used for training in the work How Should We Enhance the Safety of Large Reasoning Models: An Empirical Study. It includes data for both safety-related and mathematical reasoning tasks.
Dataset Details
Data Splits:
MATH: 4,000 mathematical reasoning examples.
Default CoT: 1,000 safety-related examples using the default CoT prompting.
RealSafe CoT: 1,000 safety-related examples with… See the full description on the dataset page: https://huggingface.co/datasets/thu-coai/LRM-Safety-Study.lrm_safety_alignment_sftifeval-lrm
IFEval Dataset Card
Dataset Description
IFEval is an instruction-following evaluation benchmark consisting of verifiable natural language instructions. Each example specifies one or more constraints that a model must satisfy in its output (e.g., include/exclude phrases, follow a format, respect length or style constraints). In this project, IFEval is used to evaluate both:
instruction following in the reasoning trace (RT), and
instruction following in the final… See the full description on the dataset page: https://huggingface.co/datasets/haritzpuerto/ifeval-lrm.lrm_safety_alignment_dpolrm_logsLRMsafetyLR-MMQAfaithful-lrm-checkpoints