Files
2024-07-04 11:34:32 +08:00

75 lines
2.6 KiB
Markdown

# Documentation
- Class name: LLMQueryEngineAsTool
- Category: SALT/Language Toolkit/Querying
- Output node: False
该节点作为一种工具,用于使用语言模型执行查询,旨在促进复杂的信息检索和处理任务。它抽象了查询语言模型的复杂性,为用户提供了一个简化的接口,以利用先进的自然语言理解能力。
# Input types
## Required
- name
- 指定工具的名称,作为查询引擎的标识符和描述符。
- Comfy dtype: STRING
- Python dtype: str
- description
- 提供工具用途和功能的详细描述,为其使用提供背景和指导。
- Comfy dtype: STRING
- Python dtype: str
- llm_index
- 标识要查询的特定索引或数据库,允许基于输入查询进行有针对性的信息检索。
- Comfy dtype: LLM_INDEX
- Python dtype: LLM_INDEX
# Output types
- query_tool
- 节点的输出,包括配置用于查询处理的工具,封装了函数及其参数。
- Comfy dtype: TOOL
- Python dtype: Dict[str, Any]
## Usage tips
- Infra type: `CPU`
- Common nodes: unknown
## Source code
```python
class LLMQueryEngineAsTool:
@classmethod
def INPUT_TYPES(cls):
return {
"required": {
"name": ("STRING", {"multiline": False, "dynamicPrompts": False, "placeholder": "code"}),
"description": ("STRING", {"multiline": True, "dynamicPrompts": False, "default": "A function that allows you to communicate with a document. Ask a question and this function will find information in the document and generate an answer."}),
"llm_index": ("LLM_INDEX",),
},
}
RETURN_TYPES = ("TOOL",)
RETURN_NAMES = ("query_tool",)
FUNCTION = "return_tool"
CATEGORY = f"{MENU_NAME}/{SUB_MENU_NAME}/Querying"
def return_tool(self, name, description, llm_index):
def query_engine(query: str) -> str:
query_components = []
query_components.append("Analyze the above document carefully to find your answer. If you can't find one, say so.")
if query:
if query.strip():
query_components.append("user: " + query)
query_components.append("assistant:")
pprint(query_components, indent=4)
query_join = "\n".join(query_components)
query_engine = llm_index.as_query_engine()
response = query_engine.query(query_join)
pprint(response, indent=4)
return (response.response,)
tool = {"name": name, "description": description, "function": query_engine}
return (tool,)
```