从社区数据看大模型开发生态的全景与趋势

Industry Investment Rating - The report does not explicitly provide an industry investment rating [1][2][3] Core Viewpoints - The report focuses on the AI development ecosystem, particularly the open-source community and its dynamics [11][12][13] - It highlights the importance of community behavior data in understanding project relationships, though it acknowledges that this data is not comprehensive [10] - The report emphasizes the potential for AI to redefine software development and the emergence of new ecosystem roles [61][64] Case Studies - The report examines the transition of popular AI projects from individual initiatives to community-managed entities, such as OpenAI/Triton to Triton-Lang/Triton and JMorganca/Ollama to Ollama/Ollama [22] - It also discusses the evolution of project names and their impact on public perception, such as LLaMA-Factory and OpenHands [22] Ecosystem Overview - The report outlines the AI ecosystem's structure, including development, application, infrastructure, and supply chain components [25][26][27] - It identifies key areas of change within the AI ecosystem, such as distributed pre-training, model size, and the shift from database-centric to model-centric services [28] Hotspots and Trends - The report tracks the development trends of traditional machine learning frameworks, such as Caffe, Keras, Caffe2, TensorFlow, and PyTorch [35][37] - It also examines the growth of large-scale distributed computing engines like Apache/Spark, Dask/Dask, and Ray-Project/Ray [38][40] - The report highlights the rise of inference deployment service engines, including HuggingFace/Text-Generation-Inference, VLLM-Project/VLLM, and NVIDIA/TensorRT-LLM [43][45] - AI application orchestration frameworks, such as LangChain-AI/LangChain, Significant-Gravitas/AutoGPT, and Microsoft/AutoGen, are also discussed [48][50] - The report notes the evolution of traditional IDEs and AI interactive development tools, including JetBrains/IntelliJ-Community, Microsoft/VSCode, and JupyterLab/JupyterLab [53][55] Architectural Insights - The report suggests that model runtime may become a first-class citizen alongside application runtime [33] - It also predicts further changes in CI and testing domains due to AI's influence [33] Strategic Recommendations - The report advises participants in the AI ecosystem to focus on trending projects and new ecosystem roles [61] - It encourages leveraging existing ecosystem positions or creating new ones for those developing their own projects [61] Conclusion - The report concludes that large language models are redefining software and bringing subtle yet positive changes to the open-source world [64]