Based on MetaMathQA and MathInstruct
, we use machine translation to compile mCoT-MATH, the first large-scale multilingual math CoT reasoning dataset containing around 6.4 million samples for 11 diverse languages.
We train a 7B parameter model mCoT for… See the full description on the dataset page:
https://huggingface.co/datasets/laihuiyuan/mCoT-MATH.