一个可以在A5000显卡或4090上完整运行的QWen3.5大模型（上下文64k），具备调用工具的能力，适合本地部署龙虾和Hermesk
tools
thinking
{
"num_ctx": 65536,
"presence_penalty": 1.5,
"repeat_last_n": 512,
"repeat_penalty": 1.15,
"stop": [
"<|im_start|>",
"<|im_end|>"
],
"temperature": 1,
"top_k": 20,
"top_p": 0.95
}