【新华社/人民日报/中央电视台/华尔街日报/纽约时报综合报道】今日，全球知名科技公司MiniMax稀宇科技发布了一项重大突破，正式推出了其万亿参数级别的大规模预训练模型——abab 6.5系列。该系列包括abab 6.5和abab 6.5s两个版本，标志着人工智能语言处理技术迈入新的里程。
据官方介绍，abab 6.5模型拥有万亿级别的参数量，能够处理长达200k tokens的上下文信息，其强大的处理能力可与当前世界领先的大语言模型，如GPT-4、Claude-3和Gemini-1.5相媲美。在各类核心能力的测试中，abab 6.5展现出卓越的性能，逐步接近这些顶级模型的水平。
此外，abab 6.5s作为该系列的高效版本，虽然同样采用了与abab 6.5相同的训练技术和数据，但在速度上更胜一筹。abab 6.5s能在1秒钟内处理将近3万字的文本，展现出惊人的实时处理效率，这将极大地提升大规模文本分析和应用的效率。
MiniMax稀宇科技的这一创新，无疑将对人工智能领域产生深远影响，为科研、教育、媒体和各行各业提供更强大、更快速的语言处理工具，进一步推动信息时代的智能化进程。公司表示，将持续致力于技术研发，以期在未来引领人工智能语言模型的发展方向。
英语如下：
**News Title:** “MiniMax Unveils MiniMax abab 6.5: A Trillion-Parameter MoE Model, Challenging the Global Peak of Language Models”
**Keywords:** MiniMax, Trillion Parameters, MoE Model
**News Content:**
**[Xinhua News Agency/People’s Daily/CCTV/Wall Street Journal/New York Times Joint Report]** Today, renowned global tech company MiniMax Xiyu Technology announced a groundbreaking advancement with the launch of its trillion-parameter-scale large pre-training model series, the abab 6.5. The series comprises two versions, abab 6.5 and abab 6.5s, signaling a new era in artificial intelligence (AI) language processing technology.
According to official statements, the abab 6.5 model boasts a parameter count in the trillions, capable of handling context information of up to 200k tokens. Its processing power rivals that of leading large language models such as GPT-4, Claude-3, and Gemini-1.5. In various core capability tests, abab 6.5 has demonstrated exceptional performance, gradually approaching the level of these top-tier models.
Moreover, the abab 6.5s, an efficient variant of the series, utilizes the same training techniques and data as abab 6.5 but excels in speed. It can process nearly 30,000 characters of text within a second, exhibiting remarkable real-time processing efficiency. This will significantly enhance the efficiency of large-scale text analysis and applications.
MiniMax Xiyu Technology’s innovation is set to have a profound impact on the AI domain, providing stronger and faster language processing tools for research, education, media, and industries, further accelerating the progression of the information age’s智能化. The company has expressed its commitment to continuous R&D, aspiring to shape the future direction of AI language models.
【来源】https://mp.weixin.qq.com/s/xBoAP-6fZVQA9cEWT8gyfw
Views: 1