LoRA adapters for XQDT trained on the joint WebNLG–E2E synthetic training set across Qwen3, Gemma 3, and Llama model families.