community-alignment-dataset

Community Alignment Github   |   Paper Dataset Community Alignment* is a large-scale open source, multilingual and multi-turn preference dataset to align LLMs with human preferences across cultures. Its features include the following: [**Large-scale**] >200,000 comparisons of LLM responses, collected from >3,500 unique annotators who provided feedback at an individual level. [**Multilingual**] Contains comparisons in English, French, Italian, Hindi, and Portuguese. 66% of comparisons are non-English. [**Prompt-level overlap**] >2500 prompts feature at least 10 annotations per prompt. This enables research on social-choice-based and distributional approaches to LLM alignment [**High-quality natural language explanations**] For 44% of comparisons, annotators provided detailed explanations why they preferred one response over another. [**Many conversations per annotator**] The median number of conversations per annotator is 26, making Community Alignment a promising dataset for exploring methods for personalization. License Community Alignment is released under the Creative Commons Attribution 4.0 International License (CC-BY-4.0). Codebook Please see Appendix I of the paper for the codebook. Usage In ~31% of the conversations in our dataset, annotators initiate the dialogue with their own prompts. These prompts do not reflect the position of Meta or its employees. Users must implement appropriate filtering and moderation measures when utilizing this dataset for training purposes to ensure that the generated outputs adhere to their own content standards. The user-initiated conversations can be easily filtered out of the dataset using the `is_pregenerated_first_prompt` flag. Version History **v1.1**: Added 41K additional comparisons from a new data-collection wave involving 407 new annotators. This wave focused on underrepresented demographic groups, and the dataset now includes balanced subsets across all five countries. The new dataset also has more natural-