model_name_or_path: Qwen/Qwen3-4B-Instruct-2507 template: qwen3_nothink infer_backend: huggingface # choices: [huggingface, vllm, sglang, ktransformers] trust_remote_code: true
