国产算力适配
Search documents
智谱发布GLM-5技术细节:工程级智能,适配国产算力
Hua Er Jie Jian Wen· 2026-02-22 11:20
Core Insights - The release of GLM-5 marks a significant advancement in AI model capabilities, shifting the focus from mere parameter size to system engineering capabilities [2][15] - GLM-5 demonstrates the ability to perform complex tasks, improve training efficiency, and fully adapt to domestic chip architectures, indicating a move towards an independent technological ecosystem in China [2][14] Group 1: Model Capabilities - GLM-5 can handle complex tasks beyond simple code generation, showcasing "engineering-level intelligence" [4][5] - The model supports a context length of 200K tokens, enabling it to manage long-term planning and multi-round interactions effectively [4][6] - The introduction of DSA (DeepSeek Sparse Attention) reduces computational complexity by 1.5-2 times without loss of performance, allowing for more efficient processing [6][7][9] Group 2: Training and Efficiency Innovations - GLM-5 features a restructured reinforcement learning (RL) architecture that decouples model generation from training, significantly enhancing throughput [13] - The model's training efficiency is optimized through asynchronous RL algorithms, allowing for stable learning in complex environments [13] - The overall design emphasizes efficiency innovations over sheer computational power, which is crucial for the Chinese AI landscape [10] Group 3: Hardware Adaptation - GLM-5 is natively compatible with various domestic GPU ecosystems, including Huawei Ascend and others, marking a shift towards system-level adaptation rather than reliance on foreign hardware [14] - The model's performance on a single domestic computing node is comparable to that of a cluster of two international GPUs, with deployment costs reduced by 50% in long-sequence processing scenarios [14] Group 4: Comprehensive AI Engineering - The development of GLM-5 represents a complete closed-loop system that integrates model architecture innovation, training efficiency optimization, and deep adaptation to domestic chips [15] - This signifies a transition for Chinese AI from application-level advantages to full-stack optimization, including architecture, algorithms, training systems, and inference frameworks [15][18] - The report emphasizes a mature approach to AI development, focusing on practical engineering metrics rather than competitive benchmarking [18]
MinerU全面深度适配主流国产算力,10余家国产AI芯片在列
Huan Qiu Wang· 2026-02-12 08:45
Group 1 - The collaboration between OpenDataLab, DeepLink, and domestic chip manufacturers has led to the adaptation of over 10 mainstream domestic computing power solutions, enhancing the ecological compatibility and adaptability of the MinerU project [1] - MinerU's self-developed VLM model achieves an accuracy rate of 99% in capturing elements from PDFs and complex web pages, enabling precise restoration and structured extraction of intricate mathematical formulas and nested structured tables [1] - The core value of MinerU lies in its cross-industry applicability and high parsing precision, serving as an efficient data production engine for large model development and a precise document parsing tool for government and enterprise sectors [1] Group 2 - TaiChuang YuanQi has completed the adaptation of over 30 AI large models to domestic computing power, including models like DeepSeek, Qianwen, and MinerU, covering various series such as Qwen3 Dense/MoE and multi-modal understanding models [2] - The ongoing updates and adaptations aim to accelerate the integration of intelligent computing with industry, enhancing the capabilities of both AI models and domestic chip manufacturers [2]
新品密集发布+国产算力适配 商汤科技股价早盘拉升8.5%
Zhi Tong Cai Jing· 2025-12-22 05:45
Core Insights - SenseTime's stock price experienced a significant increase, with a peak rise of over 8.5%, reaching HKD 2.04, driven by recent technological advancements and product innovations in AI [1] - The company launched several new AI-based products during its product release week, including the first integrated creative multi-episode generation AI agent "Seko 2.0" and the first native AI office assistant "Xiao Huan Xiong 3.0" [1] - SenseTime achieved a milestone in adapting its self-developed models to domestic chips, with a strategic partnership established with companies like Zhongke Shuguang and Daxiao for deep integration of its large models [1] Product Innovations - New AI products introduced include the "Ruying" marketing AI engine, "Daxiao Robot" for embodied intelligence, "Kapi Camera" AI photo assistant, and a new generation AI financial assistant "Kapi Accounting" [1] - The SenseCore device has completed full adaptation with mainstream domestic AI chips, including Huawei Ascend, Cambricon, and others, enhancing operational efficiency [2] Strategic Collaborations - SenseTime has established a joint optimization mechanism with chip manufacturers to significantly improve operational efficiency on domestic chips [2] - The company is positioned for potential valuation recovery as product deployment progresses and industry demand increases [2]