CoT-Self-Instruct: Building high-quality synthetic prompts for reasoning and non-reasoning tasks
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866916930620751872 |
|---|---|
| author | Yu, Ping Lanchantin, Jack Wang, Tianlu Yuan, Weizhe Golovneva, Olga Kulikov, Ilia Sukhbaatar, Sainbayar Weston, Jason Xu, Jing |
| author_facet | Yu, Ping Lanchantin, Jack Wang, Tianlu Yuan, Weizhe Golovneva, Olga Kulikov, Ilia Sukhbaatar, Sainbayar Weston, Jason Xu, Jing |
| contents | We propose CoT-Self-Instruct, a synthetic data generation method that instructs LLMs to first reason and plan via Chain-of-Thought (CoT) based on given seed tasks, and then generate a new synthetic example of similar quality and complexity. This is followed by a filtering step to select high-quality data using automatic metrics, which are then used for LLM training. In verifiable reasoning, our synthetic data significantly outperforms existing training datasets, such as s1k and OpenMathReasoning, when evaluated on MATH500, AMC23, AIME24, and GPQA-Diamond. For non-verifiable instruction-following tasks, our method surpasses the performance of both human and standard Self-Instruct training data on the AlpacaEval 2.0 and Arena-Hard benchmarks. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2507_23751 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | CoT-Self-Instruct: Building high-quality synthetic prompts for reasoning and non-reasoning tasks Yu, Ping Lanchantin, Jack Wang, Tianlu Yuan, Weizhe Golovneva, Olga Kulikov, Ilia Sukhbaatar, Sainbayar Weston, Jason Xu, Jing Artificial Intelligence Computation and Language We propose CoT-Self-Instruct, a synthetic data generation method that instructs LLMs to first reason and plan via Chain-of-Thought (CoT) based on given seed tasks, and then generate a new synthetic example of similar quality and complexity. This is followed by a filtering step to select high-quality data using automatic metrics, which are then used for LLM training. In verifiable reasoning, our synthetic data significantly outperforms existing training datasets, such as s1k and OpenMathReasoning, when evaluated on MATH500, AMC23, AIME24, and GPQA-Diamond. For non-verifiable instruction-following tasks, our method surpasses the performance of both human and standard Self-Instruct training data on the AlpacaEval 2.0 and Arena-Hard benchmarks. |
| title | CoT-Self-Instruct: Building high-quality synthetic prompts for reasoning and non-reasoning tasks |
| topic | Artificial Intelligence Computation and Language |
| url | https://arxiv.org/abs/2507.23751 |