我们可以直接像使用大模型那样使用知识库。

首先我们设置一个模板：

-- 设置模板
set template='''User: You are a helpful assistant. Think it over and answer the user question correctly. 
{context}
Please answer based on the content above? ：
{query}
Assistant:''' where scope="session";

然后我们可以使用 byzer_docs_qa 函数进行问答了：

--查看完整结果
select llm_response_predict(
byzer_docs_qa(llm_param(map(
"instruction","请使用 Byzer 加载一个csv文件",
"k",1,
"prompt","${template}",
"strategy","full_doc",
"format","json",
"temperature",0.1
)))
) as response as output;

查询常见参数：