INTELLECT-MATH-SFT-Data
- huggingface.co
Updated Jan 22, 2025
Share
Facebook
Twitter
EmailClick to copy linkLink copiedCitePrime Intellect (2025). INTELLECT-MATH-SFT-Data [Dataset]. https://huggingface.co/datasets/PrimeIntellect/INTELLECT-MATH-SFT-DataCroissantCroissant is a format for machine-learning datasets. Learn more about this at mlcommons.org/croissant.Dataset updatedJan 22, 2025Dataset provided byAuthorsPrime IntellectLicenseMIT Licensehttps://opensource.org/licenses/MIT
License information was derived automaticallyDescriptionINTELLECT-MATH: Frontier Mathematical Reasoning through Better Initializations for Reinforcement Learning
INTELLECT-MATH is a 7B parameter model optimized for mathematical reasoning. It was trained in two stages, an SFT stage, in which the model was fine-tuned on verified QwQ outputs, and an RL stage, in which the model was trained using the PRIME-RL recipe. We demonstrate that the quality of our SFT data can impact the performance and training speed of the RL stage: Due to its… See the full description on the dataset page: https://huggingface.co/datasets/PrimeIntellect/INTELLECT-MATH-SFT-Data.
NuminaMath-QwQ-CoT-5M
- huggingface.co
Updated Jan 22, 2025ShareFacebookTwitterEmailClick to copy linkLink copiedCitePrime Intellect (2025). NuminaMath-QwQ-CoT-5M [Dataset]. https://huggingface.co/datasets/PrimeIntellect/NuminaMath-QwQ-CoT-5MCroissantCroissant is a format for machine-learning datasets. Learn more about this at mlcommons.org/croissant.Dataset updatedJan 22, 2025Dataset provided byAuthorsPrime IntellectLicenseMIT Licensehttps://opensource.org/licenses/MIT
License information was derived automaticallyDescriptionINTELLECT-MATH: Frontier Mathematical Reasoning through Better Initializations for Reinforcement Learning
INTELLECT-MATH is a 7B parameter model optimized for mathematical reasoning. It was trained in two stages, an SFT stage, in which the model was fine-tuned on verified QwQ outputs, and an RL stage, in which the model was trained using the PRIME-RL recipe. We demonstrate that the quality of our SFT data can impact the performance and training speed of the RL stage: Due to its… See the full description on the dataset page: https://huggingface.co/datasets/PrimeIntellect/NuminaMath-QwQ-CoT-5M.
Not seeing a result you expected?
Learn how you can add new datasets to our index.
FacebookTwitterEmailMIT Licensehttps://opensource.org/licenses/MIT
License information was derived automatically
INTELLECT-MATH: Frontier Mathematical Reasoning through Better Initializations for Reinforcement Learning
INTELLECT-MATH is a 7B parameter model optimized for mathematical reasoning. It was trained in two stages, an SFT stage, in which the model was fine-tuned on verified QwQ outputs, and an RL stage, in which the model was trained using the PRIME-RL recipe. We demonstrate that the quality of our SFT data can impact the performance and training speed of the RL stage: Due to its… See the full description on the dataset page: https://huggingface.co/datasets/PrimeIntellect/INTELLECT-MATH-SFT-Data.