LoRA adapters for XQDT trained on WebNLG synthetic data across Qwen3, Gemma 3, and Llama model families.