开源模型

Search documents
把大模型送上天!王坚外滩大会分享:人工智能不能缺席太空
Guan Cha Zhe Wang· 2025-09-11 08:11
Core Insights - The 2025 Inclusion Bund Conference opened in Shanghai, focusing on the transformative impact of open resources in the AI era, as highlighted by Wang Jian, founder of Alibaba Cloud and director of Zhijiang Laboratory [1][5] - Wang Jian emphasized that the shift from code openness to resource openness is a revolutionary change in AI, making the choice between open and closed models a critical variable in AI competition [1][3] Group 1: AI and Open Resources - The concept of open source has evolved into open resources, where the availability of data and computational resources is essential for advancing AI [3][4] - Wang Jian compared the significance of open models in AI to the launch of the open-source browser Netscape in 1998, marking a pivotal moment in the internet era [3] Group 2: Satellite Technology and AI - In May 2023, Zhijiang Laboratory successfully launched 12 satellites, deploying an 8 billion parameter model into space, which allows for data processing directly in orbit [4] - This initiative, named the "Trisolaris Computing Constellation," aims to democratize access to satellite technology and facilitate deep space exploration by integrating AI and computational power in space [4] Group 3: Conference Overview - The 2025 Inclusion Bund Conference features a main forum, over 40 open insight forums, 18 innovation stages, and various tech-related events, emphasizing the theme of "Reshaping Innovative Growth" [5]
阿里云创始人王坚:开源与闭源模型的选择,已成为AI竞争关键变量
Xin Lang Ke Ji· 2025-09-11 02:06
Core Insights - The choice between open-source and closed-source models has become a critical variable in AI competition [1] - We are currently in an era of open-source and openness, where the openness of model weights signifies the openness of data and computing resources [1] - Merely opening software in the context of open-source is now seen as having limited impact [1]
腾讯混元最新开源成“最强翻译”:国际机器翻译比赛获30个语种第一
量子位· 2025-09-03 05:49
Core Viewpoint - Tencent's Hunyuan-MT-7B model has achieved significant success in international translation competitions, demonstrating its advanced capabilities in translating multiple languages and dialects, while also being open-sourced for broader accessibility [1][2][4]. Group 1: Model Performance and Achievements - Hunyuan-MT-7B won first place in 30 out of 31 language pairs in the WMT2025 competition, showcasing its dominance in both high-resource and low-resource languages [4][29]. - The model supports 33 languages and 5 dialects, making it a comprehensive lightweight translation solution [1]. - In the Flores200 evaluation dataset, Hunyuan-MT-7B outperformed other models of similar size and showed competitive results against larger models [6][9]. Group 2: Technical Innovations - The model is built on a complete training paradigm that includes pre-training, supervised fine-tuning, and reinforcement learning, leading to superior translation performance [11][12]. - The Shy framework, which incorporates synergy-enhanced policy optimization, fundamentally changes traditional optimization approaches by using a systematic design with two main components: foundational model development and ensemble strategies [15][19]. - The GRPO algorithm, a key innovation in the Shy framework, reduces gradient variance and improves sample efficiency, enhancing training stability and model convergence [21][24]. Group 3: Deployment and Usability - Hunyuan-MT-7B is designed for high computational efficiency, allowing for faster inference and lower operational costs compared to larger models [30]. - The model's open-source nature promotes transparency and allows for further improvements by the research community, lowering the technical barriers for participation in machine translation advancements [31]. Group 4: Broader Implications - The methodologies and frameworks developed for Hunyuan-MT-7B can serve as a reference for optimizing other specialized fields, promoting a shift from general to specialized technology applications [33].
汉王科技:公司AI电纸本上接入了DeepSeek开源模型
Mei Ri Jing Ji Xin Wen· 2025-09-02 04:21
Group 1 - The company has confirmed that it utilizes AI model technology inspired by excellent open-source models like DeepSeek for optimization [2] - The company's AI e-paper product has integrated DeepSeek's open-source model, indicating a level of collaboration [2] - Apart from the integration of DeepSeek's model, the company has not reported any other collaborations with DeepSeek [2]
任正非、梁文锋、王兴兴、彭军等入选!《时代》最新发布→
Zheng Quan Shi Bao Wang· 2025-09-01 11:52
Core Insights - The "TIME100 AI" list for 2025 has been released, featuring influential figures in the AI sector, including Chinese entrepreneurs like Ren Zhengfei from Huawei, Liang Wenfeng from DeepSeek, Wang Xingxing from Yushu Technology, and Peng Jun from Pony.ai [1][4] - The list highlights the importance of human decision-making in AI development, emphasizing that the future of technology is shaped by individuals rather than machines [1] - The presence of Chinese leaders in the list indicates that China's AI industry is emerging as a global leader in key areas such as autonomous driving, large models, and robotics [1] Company Highlights - DeepSeek, founded by Liang Wenfeng, released the DeepSeek-R1 model, which is noted for its low training cost of $6 million, challenging the necessity of large-scale projects like OpenAI's $500 billion initiative [3] - The report from Sullivan indicates that by 2025, over 80% of enterprises are expected to adopt open-source large models, driven by the performance parity between domestic and international models [3] - Pony.ai, led by Peng Jun, aims to deploy 1,000 Robotaxis by 2025, marking a significant step towards large-scale commercial operation of Level 4 autonomous driving [4] - Huawei reported a revenue of 427.04 billion yuan for the first half of 2025, a year-on-year increase of 3.95%, while net profit decreased by 32% to 37.20 billion yuan, reflecting substantial R&D investments [5] - Yushu Technology, under CEO Wang Xingxing, aims to enhance the practical value of robots in daily life, emphasizing the integration of AI and robotics for real-world problem-solving [5]
任正非、梁文锋、王兴兴、彭军等入选!《时代》最新发布→
证券时报· 2025-09-01 11:40
Core Insights - The "TIME100 AI" list for 2025 has been released, featuring influential figures in the AI sector, including notable Chinese entrepreneurs like Ren Zhengfei from Huawei and Liang Wenfeng from DeepSeek [1][5] - The list highlights the importance of human decision-making in AI development, emphasizing that the future of technology is shaped by people [1] - The presence of Chinese leaders in the list indicates China's growing competitiveness in key AI fields such as autonomous driving, large models, and robotics [1] Group 1: Key Figures and Their Contributions - Ren Zhengfei, founder of Huawei, is recognized as a pivotal leader in AI, having transformed Huawei from a small trading company into a global tech giant, now involved in cloud computing and electric vehicles [5] - Liang Wenfeng, CEO of DeepSeek, launched the DeepSeek-R1 model, which is noted for its low training cost of $6 million, challenging the necessity of large-scale projects like OpenAI's [3] - Peng Jun, CEO of Pony.ai, is the only representative from the autonomous driving sector on the list, aiming for the deployment of 1,000 Robotaxis by 2025 [4] Group 2: Market Trends and Predictions - A report by Sullivan indicates that by 2025, the performance gap between domestic open-source models and top international closed-source models will narrow, with over 80% of enterprises expected to adopt open-source large models [4] - Huawei reported a revenue of 427.04 billion yuan in the first half of 2025, a 3.95% increase year-on-year, while net profit decreased by 32% to 37.195 billion yuan, reflecting significant R&D investments [5] - Wang Xingxing, CEO of Yushutech, emphasizes the practical value of robots in daily life, highlighting the integration of AI and robotics for real-world problem-solving [6]
中国企业调用大模型日均超10万亿Tokens 阿里通义份额第一
Zheng Quan Ri Bao Wang· 2025-09-01 06:11
Core Insights - The report by Frost & Sullivan indicates a significant surge in the usage of enterprise-level large models in China, with a projected 363% increase in daily usage by the first half of 2025 compared to the end of 2024, currently exceeding 10 trillion tokens [1] Group 1: Market Growth - The enterprise-level large model usage in China is expected to experience explosive growth, with a daily average usage projected to reach 10 trillion tokens by 2025 [1] - The report highlights that Alibaba Tongyi holds the largest market share at 17.7%, making it the most chosen large model by Chinese enterprises [1] Group 2: Open Source Models - The report anticipates that as domestic models like Qwen and DeepSeek continue to be open-sourced in 2025, the performance gap between open-source models and top international closed-source models will nearly close [1] - It is projected that over 80% of enterprises will adopt open-source large models in the future, indicating that open-source models will drive a new wave of growth in the enterprise market [1]
企业级大模型报告:阿里通义第一
Yang Zi Wan Bao Wang· 2025-09-01 04:59
Core Insights - The report by Frost & Sullivan indicates that the Chinese enterprise-level generative AI market is experiencing explosive growth, with a projected daily consumption of 10.2 trillion tokens by the first half of 2025, marking a 363% increase from the second half of 2024 [1][2] Group 1: Market Overview - The daily consumption of enterprise-level large models in China is expected to exceed 10 trillion tokens by mid-2025, with Alibaba Tongyi leading the market with a 17.7% share [1] - The top three players in the market, Alibaba Tongyi, ByteDance Doubao, and DeepSeek, collectively hold over 40% of the market share [1] Group 2: Deployment Trends - 70% of enterprises are opting for public cloud deployment or utilization of large models, with 71% indicating plans to increase their use of generative AI services in public cloud formats [2] - There is a shift from seeking the "strongest single model" to finding the "optimal solution for specific business scenarios," indicating a growing demand for tailored models [2] Group 3: Open Source Models - Open source models are becoming a key growth driver in the enterprise-level large model market, with predictions that over 80% of enterprises will adopt open source large models in the future [2] - The performance gap between domestic open source models and top international closed-source models is narrowing, with models like Qwen and DeepSeek leading the way [2] Group 4: Alibaba's Developments - Alibaba Tongyi has recently open-sourced several new foundational models, including Qwen3-Coder and Qwen-Image, leading to a surge in global interest in Chinese models [3] - Alibaba has open-sourced over 300 models, establishing itself as a leader in the global open-source model market, surpassing competitors like OpenAI and Llama [3] - The Qwen3-Coder model saw a dramatic increase in usage, with a 1474% rise in one week, making it the second most used model in the programming field globally [3]
中国企业调用大模型日均超10万亿Tokens
Zheng Quan Shi Bao Wang· 2025-09-01 03:31
人民财讯9月1日电,9月1日,国际市场调研机构沙利文(Frost&Sullivan)发布了最新的《中国GenAI市场 洞察:企业级大模型调用全景研究,2025》显示,中国企业级大模型调用呈爆发式增长,2025年上半年 日均调用量较2024年底实现363%的增长,目前超10万亿Tokens。其中,阿里通义占比17.7%位列第一。 沙利文报告认为,预计未来超过80%的企业将采用开源大模型,预示着开源模型将驱动企业级市场的新 一轮增长。 ...
路线图出炉!未来十年,AI改变中国
Hua Xia Shi Bao· 2025-08-30 09:44
Group 1 - The State Council released an opinion on August 26 to implement the "Artificial Intelligence +" initiative, aiming to deeply integrate AI with various sectors and reshape production and living paradigms [1][2] - The opinion outlines a clear roadmap for AI development in China over the next decade, with goals for 2027, 2030, and 2035, including achieving over 70% penetration of new intelligent terminals and agents by 2027 [2][3] - Key actions and foundational support capabilities are detailed in the opinion, focusing on technology, industry development, consumer quality, public welfare, governance, and global cooperation [4][6] Group 2 - Companies like MiniMax and Jieyue Xingchen expressed strong support for the opinion, indicating their commitment to leveraging AI in various industries, with MiniMax planning to provide enterprise-level services [3][5] - The opinion emphasizes the importance of AI in enhancing public welfare, particularly in healthcare, with a focus on improving grassroots medical services through AI applications [5][6] - The document highlights the need for a robust action framework to support the ambitious goals set forth, including the development of a secure and controllable AI ecosystem [6][7]