Dolci-Think-SFT-32B-Multilingual: Long Chain-of-Thought Reasoning Corpus in Six Languages

Dolci-Think-SFT-32B-Multilingual is a large-scale multilingual corpus for long chain-of-thought reasoning, created by lightonai and released in 2026. It spans six languages: English, French, German, Spanish, Chinese, and Swahili. Each sample includes a question, a long-form reasoning trace, and a final answer, with sequences up to 32,768 tokens.

Use Cases

• Fine-tuning language models for multilingual reasoning based on the provided long-form reasoning traces.
• Evaluating the multilingual reasoning gap in language models based on the parallel question-answer-reasoning structure.
• Training models for complex question answering in low-resource languages like Swahili based on the translated samples.
• Studying the transfer of reasoning skills across languages using the aligned multilingual corpus.

Strengths

• Includes samples in six languages, covering major and a lower-resource language (Swahili).
• Contains long-form reasoning traces, with sequences up to 32,768 tokens.
• Released alongside a research paper ('Rethinking the Multilingual Reasoning Gap with Layer Swap'), suggesting academic rigor.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Last updated 2026-05-27 09:24:00; freshness should be verified.

Provenance

Source

lightonai

Collection Method

Likely constructed for research, with questions, reasoning traces, and answers translated into target languages.

Freshness

Last updated 2026-05-27 09:24:00.