scope
Datasets
All datasets matching “scope”SCOPE-OOD-set
SCOPE-60K-OOD: Out-of-Distribution LLM Routing Dataset
Dataset Description
SCOPE-60K-OOD is an out-of-distribution (OOD) evaluation dataset for LLM routing systems. It contains evaluation results from 5 frontier language models that were not seen during training, designed to test the generalization capabilities of routing methods.
Authors
Qi Cao - UC San Diego, PXie Lab
Shuhao Zhang - UC San Diego, PXie Lab
Affiliation
University of California, San… See the full description on the dataset page: https://huggingface.co/datasets/Cooolder/SCOPE-OOD-set.SCoPE-Gallery
SCoPE Project Gallery
This dataset contains the 182 generated result videos used by the project page for SCoPE: Sightline-Coordinate Positional Encoding for Video Diffusion Transformers. Each clip is generated from a single input frame and a target camera trajectory. The picture-in-picture overlay visualizes the commanded camera motion.
Links
Project page
Paper
Code
Model
Structure
The MP4 files remain individually addressable so the project page… See the full description on the dataset page: https://huggingface.co/datasets/Michaelqaz/SCoPE-Gallery.scopebench-pilot
ScopeBench pilot trajectories
This dataset contains the 2,160 ATIF trajectories produced for
ScopeBench: Do Agents Preserve Engagement Boundaries Under Goal Pressure? at AISec 2026. The
corresponding frozen tasks, evaluation runner, and verifiers are available in the
dreadnode/scopebench-pilot GitHub repository.
Dataset structure
The pilot crosses 30 tasks, three instruction conditions, eight acting-model families, and three
repetitions. Each JSONL file contains… See the full description on the dataset page: https://huggingface.co/datasets/dreadnode/scopebench-pilot.SCoPE
SCoPE Dataset
Dataset · Code · Paper
SCoPE contains 189 images and 756 image–category questions for evaluating controllable image captioning. Each image is paired with four semantic focuses: Attribute, Relation, Foreground, and Background.
Download
From the cloned GitHub repository, install the dependencies and download the benchmark:
hf download mldljyh/SCoPE \
--repo-type dataset \
--local-dir . \
--include "benchmark_images_final/*" \
--include… See the full description on the dataset page: https://huggingface.co/datasets/mldljyh/SCoPE.ScopeBOscopes_test
Dataset Card for "scopes_test"
More Information needed