Maximizing Capacity Utilization of LoRA Adapters for Mitigating the Learning-Forgetting Trade-off — ThinkLLM