参考
https://docs.vllm.ai/en/latest/features/reasoning_outputs.html
和
https://github.com/deepseek-ai/DeepSeek-R1/issues/352
1.修改模型的tokenizer_config.json 去掉 最后面的 <think>//n
2.vllm部署时添加命令–enable-reasoning --reasoning-parser deepseek_r1
3.maxKB提示词强制开启思考
