README / README.md
R1ckyTu's picture
Rename: CLE-Lite (cards), CLE-Lite-Workspaces, CLE-Lite-Rubrics, CLE-Lite-Full; update cross-links
5ff747e verified
|
Raw History Blame Contribute Delete
1.88 kB
---
title: README
emoji: 📝
colorFrom: indigo
colorTo: blue
sdk: static
pinned: false
---
# Continual Learning Exam
Continual Learning Exam studies long-horizon agent work in professional settings. CLE Lite contains twenty tasks, each a five-level job for one real occupation. New material arrives as the work progresses, and earlier deliverables carry forward.
**[Project site](https://haoyu-zhao.github.io/CLE/)**
## How to use the datasets
These four repositories describe **the same twenty tasks**, with different kinds of material and different access levels.
| If you want to… | Dataset | What it contains | Access |
|---|---|---|---|
| Read the assignments | [CLE Lite](https://huggingface.co/datasets/Continual-Learning-Exam/CLE-Lite) | A browsable table of 100 level records: instructions, requested outputs, and an inventory of the input files arriving at each level. | Public |
| Inspect the material given to an agent | [CLE Lite Workspaces](https://huggingface.co/datasets/Continual-Learning-Exam/CLE-Lite-Workspaces) | Instructions and task-facing files, including messages, documents, data, and source notices. | Public |
| Review how work is judged | [CLE Lite Rubrics](https://huggingface.co/datasets/Continual-Learning-Exam/CLE-Lite-Rubrics) | Evaluation rubrics and ground-truth references, kept separate from the agent's inputs. | Manual approval required to download |
| Access the complete task packages | [CLE Lite Full](https://huggingface.co/datasets/Continual-Learning-Exam/CLE-Lite-Full) | The Harbor tasks and CLE packages with inputs, environment and setup material, and evaluation material. | Manual approval required to download |
The two approval-gated repositories have visible pages, but their files cannot be downloaded without approval. The public CLE Lite cards and workspaces do not include reference answers or evaluation code.