Community Alignment: Multilingual Human Preference Comparisons for LLMs

Over 200,000 comparisons of large language model responses were collected from more than 3,500 unique annotators. The dataset is multilingual, containing comparisons in English, French, Italian, Hindi, and Portuguese. It was created by Facebook and last updated on Hugging Face in February 2026.

Use Cases

• Training reward models based on large-scale human preference comparisons.
• Fine-tuning LLMs for cross-cultural alignment using multilingual feedback.
• Analyzing annotator-level preference patterns for bias or consistency studies.
• Benchmarking LLM response quality across different languages.

Strengths

• Large scale with over 200,000 preference comparisons.
• Multilingual coverage across five languages.
• Feedback collected from a diverse pool of over 3,500 individual annotators.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect geographic or cultural bias inherent to the annotator pool.

Provenance

Source

Facebook

Collection Method

Likely contains comparisons of LLM responses collected from individual annotators providing feedback.

Time Range

null

Freshness

Last updated 2026-02-19 22:30:37; freshness should be verified.

Geography

null