Macaron-V1：面向开放持续学习的自改进与LoRA混合架构
原标题：Macaron-V1: Towards Open Continual Learning with Self-Improvement and Mixture-of-LoRA
AI 摘要
Macaron-V1 是一个面向开放持续学习的智能体模型系列，通过递归改进版本化模型-工具对实现自适应，并采用 Mixture-of-LoRA 架构冻结基础模型、组合专家 LoRA 适配器。旗舰版 Macaron-V1-Venti 基于 744B GLM-5.2 基础模型，配备四个 LoRA 模块，而 Macaron-V1-Tall 基于 Qwen3.6 用于本地部署。该系列在个人智能、GenUI 和通用能力基准上进行了评估，验证了当前系统，但持续学习和集体智能的复合收益仍是开放问题。
正文节选
Macaron-V1: Towards Open Continual Learning with Self-Improvement and Mixture-of-LoRA Abstract Macaron-V1 is an open agent-model family for experiential intelligence: learning from experience in real environments and continuing to learn after deployment. It is organized around two system goals. Adaptation is pursued through recursive improvement of versioned model-harness pairs, where experience from one configuration is evaluated under an external contract and used to construct its successor. C