AB
AB
AiBoss

All Categories

project

MAI-DS-R1 - Microsoft's open-source AI model, an improved version of DeepSeek R1.

MAI-DS-R1 is an AI model developed by Microsoft based on DeepSeek R1. MAI-DS-R1 is post-training optimized and supports responding to 99.3% of sensitive topic prompts, a 2x improvement over the original version, reducing the risk of harmful content by 50%. MAI...

What is MAI-DS-R1?

MAI-DS-R1 is an AI model developed by Microsoft based on DeepSeek R1. Optimized through post-training, MAI-DS-R1 supports 99.3% of sensitive topic prompts, a 2x improvement over the original version, and reduces the risk of harmful content by 50%. MAI-DS-R1 maintains the same level of inference ability as DeepSeek R1, supports multilingual responses, and is suitable for multilingual environments such as international organizations, multinational corporations, and educational institutions. MAI-DS-R1 is open source and available for researchers and developers.

Main functions of MAI-DS-R1

  • Efficient response to sensitive topicsIt supports responding to 99.3% of sensitive questions, significantly outperforming the original DeepSeek R1.
  • Low riskIn the security assessment, the risk of harmful content was reduced by 50%.
  • reasoning abilityIt maintains the same reasoning capabilities as DeepSeek R1, making it suitable for complex logic and knowledge-based problems.
  • Multilingual supportSupports multiple languages to adapt to different language environments.

Technical Principles of MAI-DS-R1

  • Post-trainingThis study optimizes the original DeepSeek R1 model using post-training techniques. Post-training involves fine-tuning the model with specific datasets and strategies after pre-training to improve its performance on specific tasks. Microsoft used approximately 350,000 examples of blocked topics for post-training, covering a variety of sensitive topics. MAI-DS-R1 learned how to respond more effectively to these topics, avoiding the generation of harmful content.
  • Data AugmentationDuring post-training, Microsoft added 110,000 security and violation examples from the Tulu3 SFT dataset, including examples on sensitive topics. These examples included content from CoCoNot, WildJailbreak, and WildGuardMix, helping the model better identify and handle potentially harmful content.
  • Multilingual translationDuring post-training, the problem is translated into multiple languages to adapt to the needs of different linguistic environments. This improves the model's multilingual capabilities and allows it to better understand problems from different cultural backgrounds.
  • Security assessmentMicrosoft conducted a comprehensive security assessment of MAI-DS-R1, using the HarmBench dataset to detect harmful content generated by the model and ensure that the output complies with ethical and legal standards.

MAI-DS-R1 project address

Application scenarios of MAI-DS-R1

  • academic researchIt helps researchers quickly acquire and organize multi-faceted information on sensitive topics, assists in writing academic papers, and provides more comprehensive discussion content.
  • Content moderationUsed on social media and news platforms to efficiently identify and filter harmful or inappropriate information, ensuring the health and safety of content.
  • Multilingual customer serviceProvide multilingual support for multinational corporations or international organizations, quickly respond to inquiries from users in different languages, and improve customer service efficiency and user experience.
  • Educational guidanceIn educational institutions, it assists teachers in their teaching, provides students with multilingual academic guidance and problem-solving, and promotes the dissemination of knowledge.
  • Policy ConsultationTo provide data support and public opinion analysis for government agencies or policy research institutions to analyze sensitive social issues and assist in the formulation of more reasonable policies.
Back