Tag
This paper introduces Macaron-V1, an open continual learning agent-model family using Mixture-of-LoRA to compose specialist adapters on frozen base models, with recursive self-improvement and model-harness co-design.