Mathoctopus/Cross_xRFT_7B

TEXT GENERATIONConcurrency Cost:1Model Size:7BQuant:FP8Ctx Length:4kLicense:apache-2.0Architecture:Transformer0.0K Open Weights Cold

Mathoctopus/Cross_xRFT_7B is a 7 billion parameter LLaMA 2-based large language model developed by MathOctopus, specifically fine-tuned for multilingual mathematical reasoning. This model is trained using a cross-training strategy with multilingual rejection sampling (xRFT) on the MGSM8KInstruct Dataset, which covers ten languages. It demonstrates enhanced performance in solving math problems across multiple languages, outperforming conventional open-source LLMs and showing superiority over ChatGPT in few-shot scenarios for mathematical tasks.

Loading preview...

Mathoctopus/Cross_xRFT_7B: Multilingual Mathematical Reasoning

Mathoctopus/Cross_xRFT_7B is a 7 billion parameter model from the MathOctopus series, built upon the LLaMA 2 architecture. It is specifically designed and optimized for solving mathematical problems across multiple languages. The model was trained using a unique cross-training strategy combined with multilingual rejection sampling (xRFT), leveraging the extensive šŸ¤— MGSM8KInstruct Dataset which includes over 73,000 math problems across ten distinct languages (English, Swahili, Chinese, Bengali, German, Spanish, French, Japanese, Russian, Thai).

Key Capabilities

  • Multilingual Math Problem Solving: Excels at mathematical reasoning in 10 different languages.
  • Enhanced Performance: Outperforms conventional open-source LLMs and demonstrates superiority over ChatGPT in few-shot mathematical reasoning tasks.
  • Rejection Sampling: Utilizes multilingual rejection sampling (xRFT) during training to improve accuracy.
  • Robust Training Data: Trained on the comprehensive MGSM8KInstruct Dataset and evaluated on MSVAMP, ensuring broad language coverage and mathematical complexity.

Good For

  • Educational Software: Developing applications for teaching and practicing math in various languages.
  • Tutoring Systems: Creating AI-powered tutors capable of assisting students with math problems globally.
  • Research: Investigating multilingual reasoning capabilities in LLMs and exploring advanced training strategies like xRFT.
  • Applications requiring math problem solutions: Any system where accurate, multilingual mathematical problem-solving is a core requirement.

Finetunes
1 models