HelpSteer2: Open-source Dataset for Training Top-Performing Reward Models
NVIDIA-released 10K-sample multi-attribute preference dataset (helpfulness, correctness, coherence, complexity, verbosity) for training reward models.
- Publisher
- NVIDIA
- Year
- 2024
- Venue
- NeurIPS
- Authors
- 9
- Hosting
- External sourcelicense unknown
Cite
Notes
Only stored in your browser.
Introduces 1 artifact - 1 tool
TL;DR
Semantic Scholar
This work proposes SteerLM 2.0, a model alignment approach that can effectively make use of the rich multi-attribute score predicted by the reward models, and releases HelpSteer2, a permissively licensed preference dataset (CC-BY-4.0).
Artifacts
1Tools