Abstract: While early AutoML frameworks focused on optimizing traditional ML pipelines and their hyperparameters, a recent trend in AutoML is to focus on neural architecture search. In this paper, we ...

编程世界正在经历一场前所未有的变革。过去，程序员需要花费大量时间阅读成千上万行代码，理解复杂的软件架构，然后小心翼翼地修改几行关键代码。现在，AI开始能够独立完成这些工作了。Meta公司刚刚发布的这个名为"Confucius Code ...

Security researchers uncovered a range of cyber issues targeting AI systems that users and developers should be aware of — ...

在本文里，将在AndesAIRE平台上实践自带模型（BYOM）流程，它是从模型开发、ONNX导出、API设计、交叉编译，到在RISC-V上部署与推论的完整流程。此流程展示了AndesAIRE I370平台在灵活整合外部模型与支持多元运行环境的优势。

Overview:  Reinforcement learning in 2025 is more practical than ever, with Python libraries evolving to support real-world simulations, robotics, and deci ...

JAX is one of the fastest-growing tools in machine learning, and this video breaks it down in just 100 seconds. We explain ...

其核心逻辑简单而直接：通过自动化移植工具，将原本绑定在国际主流平台上的C++源代码，转化为MUSA架构的C++源代码，让开发者以最小成本将国际主流GPU平台应用移植至MUSA GPU，最终运行在全功能GPU上。

Overview: Python supports every stage of data science from raw data to deployed systemsLibraries like NumPy and Pandas simplify data handling and analysisPython ...

Thinking Machines Lab Inc. today launched its Tinker artificial intelligence fine-tuning service into general availability.

Stanford researchers have shown that AI models can now write CUDA code that runs faster than PyTorch's built-in functions in several tests. The team got better results by letting the AI try out many ...

【导读】Meta推出KernelLLM，这个基于Llama 3.1微调的8B模型，竟能将PyTorch代码自动转换为高效Triton GPU内核。实测数据显示，它的单次推理性能超越GPT-4o和DeepSeek V3，多次生成时得分飙升。