ModelScope中错误怎么解决？

https://www.modelscope.cn/docs/VLLM%E6%8E%A8%E7%90%86%E5%8A%A0%E9%80%9F%E4%B8%8E%E9%83%A8%E7%BD%B2 执行：CUDA_VISIBLE_DEVICES=0 swift infer --model_type qwen-7b-chat --infer_backend vllm 报错：ValueError: The model's max seq len (32768) is larger than the maximum number of tokens that can be stored in KV cache (21568). Try increasing gpu_memory_utilization or decreasing max_model_len when initializing the engine. ModelScope中错误怎么解决？