A Unified Moral-Value Dataset for Instruction Tuning
Merged summary
TL;DR - This paper introduces a unified moral-value dataset formatted for instruction tuning, aiming to support research on aligning LLMs with human values. Preliminary experiments suggest it can be mixed with general-task data without degrading general performance.
- Consolidates existing moral-value datasets into a single instruction-response corpus.
- Tests joint training with general-purpose instruction datasets.
- Reports preliminary evidence that value-task performance depends on the moral-data mixing ratio.
- Publicly releases the dataset on Hugging Face for further alignment research.
Sources (1)
A Unified Moral-Value Dataset for Instruction Tuning
TL;DR - This paper introduces a unified moral-value dataset formatted for instruction tuning, aiming to support research on aligning LLMs with human values. Preliminary experiments suggest it can be mixed with general-task data without degrading general performance.
- Consolidates existing moral-value datasets into a single instruction-response corpus.
- Tests joint training with general-purpose instruction datasets.
- Reports preliminary evidence that value-task performance depends on the moral-data mixing ratio.
- Publicly releases the dataset on Hugging Face for further alignment research.