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