Dataset · sojuL
RubricHub: rubrics for generation evaluation
sojuL/RubricHub_v1
About 110k examples with rubrics for training and evaluating open-ended generation in zh/en.
What's inside
A large text dataset of approximately 110K examples with rubric supervision for open-ended generation tasks. Collected via an automatic coarse-to-fine framework: principle-based synthesis, multi-model aggregation, and complexification. Languages are Chinese and English.
What you can do with it
- Training rubric-based reward models
- RLHF/RL for open-ended generation
- Evaluating LLM response quality using rubrics
- Tasks
- Text generation, Reinforcement learning, Question answering
- Languages
- Chinese, English
- Size
- 100K–1M rows
- Downloads this month
- 1.6K
- Likes
- 283
Models trained on it
How to load it
from datasets import load_dataset
ds = load_dataset("sojuL/RubricHub_v1")