npcpy
npcpy是一个面向AI Agent开发的开源框架，旨在帮助开发者构建和管理基于大语言模型的智能代理应用。该框架采用分层解耦架构设计，将不同功能模块分离，以提升系统的可维护性与扩展性，适用于需要多智能体协作或复杂任务编排的场景。
All Facts (13)
npcpy已开源，可通过pip install npcpy --upgrade安装
95%Unverifiednpcpy的NPC层将人格定义、行为指令和能力接口设计为可插拔模块
90%Unverifiednpcpy兼容Ubuntu生产环境、Mac Apple Silicon和Windows WSL
85%Unverifiednpcpy深度适配Apple Silicon，在MLX架构下运行LoRA，rank拉到128，配合Flash Attention 2.0
80%Unverifiednpcpy的模型微调使用Beta为0.1的DPO，并用SFT迁移了Llama 3的学术风格用于科学写作场景
80%Unverifiednpcpy's adversarial collaboration (iterative game) mechanism improved the resolution rate of complex problems by 42%.
60%Unverifiednpcpy's NPC Array implements vectorized parallel inference with reference to the SIMD approach, enabling the parallel processing of inference tasks for hundreds of agents.
60%Unverifiednpcpy supports model adaptation ranging from local Ollama to cloud-based OpenAI, Anthropic, and DeepSeek.
60%Unverifiednpcpy's Tool layer connects to external interfaces via the MCP protocol and supports services such as Stable Diffusion and Hugging Face.
60%Unverifiednpcpy adopts a four-layer architecture design of NPC, Context, Agent, and Tool.
60%Unverifiednpcpy的C-Memolution种群进化机制中最佳个体适应度达到91.5%，多样性指数为0.62
55%Unverifiednpcpy的睡眠机制冗余修剪率达到78%
55%Unverifiednpcpy的梦境机制推测性连接点突破了182%
50%