Mathoctopus/Cross_7B

Mathoctopus/Cross_7B: Multilingual Mathematical Reasoning

Mathoctopus/Cross_7B is a 7 billion parameter LLaMA 2-based model from the MathOctopus series, specifically designed for advanced multilingual mathematical problem-solving. Developed by Mathoctopus, this model leverages a unique cross-training strategy on the extensive MGSM8KInstruct Dataset, which covers ten languages including English, Swahili, Chinese, Bengali, German, Spanish, French, Japanese, Russian, and Thai.

Key Capabilities

• Multilingual Math Reasoning: Excels at solving mathematical problems across a diverse set of ten languages, demonstrating robust cross-lingual transfer capabilities.
• Performance: Outperforms many conventional open-source LLMs and shows superiority over ChatGPT in few-shot mathematical reasoning tasks, particularly on benchmarks like MGSM and MSVAMP.
• Cross-Training Strategy: Utilizes a 'Cross-Training' approach, distinct from 'Parallel-Training', which contributes to its strong multilingual performance.

Good For

• Research in Multilingual AI: Ideal for researchers exploring cross-lingual transfer and mathematical reasoning in LLMs.
• Educational Software: Suitable for integration into educational applications requiring accurate math problem-solving in various languages.
• Tutoring Systems: Can power intelligent tutoring systems that assist users with mathematical challenges across different linguistic backgrounds.
• Benchmarking: Useful for evaluating and comparing multilingual mathematical reasoning capabilities against other models.