datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
unkebench-hpse
UnKEBench-HPSE
This repository contains the UnKEBench evaluation data used in
Hybrid-Policy Self-Editing for Composable Unstructured Knowledge Editing.
It extends the 1,000 records in UnKEBench with an untargeted editing prompt for each passage.
The project repository, including lightweight evaluation helpers, is available at
lliutianc/hpse.
Usage
from datasets import load_dataset
dataset = load_dataset("lliutianc/unkebench-hpse", split="test")
print(dataset[0])… See the full description on the dataset page: https://huggingface.co/datasets/lliutianc/unkebench-hpse.CHaMEL_v1.0
CHaMEL v1.0 Dataset
Controllable Harness for Model adaptability Evaluation under Latent rule-shifts
Dataset Description
CHaMEL is a benchmark harness for evaluating the adaptive reasoning capabilities of Large Language Models (LLMs) through dynamic rule-shift tasks. This dataset contains the pre-generated task stimuli and human evaluation baselines used in the CHaMEL paper.
CHaMEL implements a factorial experiment design where researchers can independently control the… See the full description on the dataset page: https://huggingface.co/datasets/unknown202612/CHaMEL_v1.0.