新华社北京讯，今日，全球知名科技公司 MiniMax 稀宇科技宣布了一项重大创新，正式推出其最新一代的超大规模语言模型——abab 6.5 系列。该系列包括两个模型：abab 6.5 和 abab 6.5s，标志着人工智能语言处理技术的又一里程碑。
abab 6.5 模型引人注目，拥有万亿级别的参数量，能够处理长达 200,000 个令牌（tokens）的上下文信息，这在当前的语境理解能力测试中，已开始显示出与世界领先的 GPT-4、Claude-3 和 Gemini-1.5 等大语言模型相媲美的潜力。
同时推出的 abab 6.5s 模型，虽然同样基于先进的训练技术和海量数据，但在效率上更胜一筹。它能够在短短一秒钟内处理近 30,000 字的文本，展现出强大的实时处理能力。这一速度优势使得 abab 6.5s 在高负荷的文本处理场景中具有显著的竞争优势。
MiniMax 稀宇科技的这一创新发布，不仅在技术层面刷新了行业标准，也为人工智能在文本理解和生成领域开辟了新的可能。据公司发言人表示，abab 6.5 系列模型将广泛应用于自然语言处理、机器翻译、智能客服以及内容创作等多个领域，有望引领新一轮的技术革新。
此次 MiniMax 的万亿 MoE 模型发布，无疑为全球科技界带来了一场震撼，预示着人工智能在理解复杂语言结构和大规模文本处理方面取得了重大突破，进一步推动了人机交互的智能化进程。
英语如下：
**News Title:** “MiniMax Sets a Record with the Trillion-Parameter MoE Model abab 6.5, Challenging the Peak of Global Language Models”
**Keywords:** MiniMax, trillion parameters, MoE model
**News Content:**
**Beijing, Xinhua News** – MiniMax Xiyu Technology, a globally renowned tech company, has announced a groundbreaking innovation today by launching its latest generation of ultra-large language models – the abab 6.5 series. This series comprises two models: abab 6.5 and abab 6.5s, marking another milestone in artificial intelligence language processing.
The abab 6.5 model, with its impressive trillion parameters, is capable of handling contextual information of up to 200,000 tokens, demonstrating potential comparable to leading models like GPT-4, Claude-3, and Gemini-1.5 in current benchmark tests for contextual understanding.
The concurrently launched abab 6.5s, while also leveraging advanced training techniques and vast amounts of data, excels in efficiency. It can process nearly 30,000 characters of text in just one second, showcasing its superior real-time processing capabilities. This speed advantage gives abab 6.5s a significant edge in high-volume text processing scenarios.
By pushing the boundaries in the industry with this innovation, MiniMax Xiyu Technology has not only set new technical standards but also opened up new possibilities in text understanding and generation. According to a company spokesperson, the abab 6.5 series will be widely applied in natural language processing, machine translation, intelligent customer service, and content creation, potentially spearheading a new wave of technological advancements.
MiniMax’s introduction of the trillion-parameter MoE model has undoubtedly sent shockwaves through the global tech community, signifying a major breakthrough in understanding complex language structures and handling large-scale text processing. This development further propels the progression towards smarter human-computer interaction.
【来源】https://mp.weixin.qq.com/s/xBoAP-6fZVQA9cEWT8gyfw
Views: 1