mirror of
https://github.com/langgenius/dify-docs.git
synced 2026-03-27 13:28:32 +07:00
295 lines
8.2 KiB
Plaintext
295 lines
8.2 KiB
Plaintext
---
|
||
dimensions:
|
||
type:
|
||
primary: implementation
|
||
detail: advanced
|
||
level: intermediate
|
||
standard_title: Reverse Invocation Model
|
||
language: zh
|
||
title: Model
|
||
description: 本文档详细介绍了插件如何反向调用Dify平台中的模型服务。内容包括反向调用LLM、Summary、TextEmbedding、Rerank、TTS、Speech2Text和Moderation等模型的具体方法，每种模型调用都配有对应的入口、接口参数说明以及实际的使用示例代码，并提供了调用模型的最佳实践建议。
|
||
---
|
||
|
||
反向调用 Model 指的是插件能够反向调用 Dify 内 LLM 的能力，包括平台内的所有模型类型与功能，例如 TTS、Rerank 等。如果你对反向调用的基本概念还不熟悉，请先阅读[反向调用 Dify 服务](/plugin-dev-zh/9241-reverse-invocation)。
|
||
|
||
不过请注意，调用模型需要传入一个 `ModelConfig` 类型的参数，它的结构可以参考 [通用规范定义](/plugin-dev-zh/0411-general-specifications)，并且对于不同类型的模型，该结构会存在细微的差别。
|
||
|
||
例如对于 `LLM` 类型的模型，还需要包含 `completion_params` 与 `mode` 参数，你可以手动构建该结构，或者使用 `model-selector` 类型的参数或配置。
|
||
|
||
### 调用 LLM
|
||
|
||
#### **入口**
|
||
|
||
```python
|
||
self.session.model.llm
|
||
```
|
||
|
||
#### **端点**
|
||
|
||
```python
|
||
def invoke(
|
||
self,
|
||
model_config: LLMModelConfig,
|
||
prompt_messages: list[PromptMessage],
|
||
tools: list[PromptMessageTool] | None = None,
|
||
stop: list[str] | None = None,
|
||
stream: bool = True,
|
||
) -> Generator[LLMResultChunk, None, None] | LLMResult:
|
||
pass
|
||
```
|
||
|
||
请注意，如果你调用的模型不具备 `tool_call` 的能力，那么此处传入的 `tools` 将不会生效。
|
||
|
||
#### **用例**
|
||
|
||
如果想在 `Tool` 中调用 `OpenAI` 的 `gpt-4o-mini` 模型，请参考以下示例代码：
|
||
|
||
```python
|
||
from collections.abc import Generator
|
||
from typing import Any
|
||
|
||
from dify_plugin import Tool
|
||
from dify_plugin.entities.model.llm import LLMModelConfig
|
||
from dify_plugin.entities.tool import ToolInvokeMessage
|
||
from dify_plugin.entities.model.message import SystemPromptMessage, UserPromptMessage
|
||
|
||
class LLMTool(Tool):
|
||
def _invoke(self, tool_parameters: dict[str, Any]) -> Generator[ToolInvokeMessage]:
|
||
response = self.session.model.llm.invoke(
|
||
model_config=LLMModelConfig(
|
||
provider='openai',
|
||
model='gpt-4o-mini',
|
||
mode='chat',
|
||
completion_params={}
|
||
),
|
||
prompt_messages=[
|
||
SystemPromptMessage(
|
||
content='you are a helpful assistant'
|
||
),
|
||
UserPromptMessage(
|
||
content=tool_parameters.get('query')
|
||
)
|
||
],
|
||
stream=True
|
||
)
|
||
|
||
for chunk in response:
|
||
if chunk.delta.message:
|
||
assert isinstance(chunk.delta.message.content, str)
|
||
yield self.create_text_message(text=chunk.delta.message.content)
|
||
```
|
||
|
||
可以留意到代码中传入了 `tool_parameters` 中的 `query` 参数。
|
||
|
||
### **最佳实践**
|
||
|
||
并不建议手动来构建 `LLMModelConfig`，而是允许用户可以在 UI 上选择自己想使用的模型，在这种情况下可以修改一下工具的参数列表，按照如下配置，添加一个 `model` 参数：
|
||
|
||
```yaml
|
||
identity:
|
||
name: llm
|
||
author: Dify
|
||
label:
|
||
en_US: LLM
|
||
zh_Hans: LLM
|
||
pt_BR: LLM
|
||
description:
|
||
human:
|
||
en_US: A tool for invoking a large language model
|
||
zh_Hans: 用于调用大型语言模型的工具
|
||
pt_BR: A tool for invoking a large language model
|
||
llm: A tool for invoking a large language model
|
||
parameters:
|
||
- name: prompt
|
||
type: string
|
||
required: true
|
||
label:
|
||
en_US: Prompt string
|
||
zh_Hans: 提示字符串
|
||
pt_BR: Prompt string
|
||
human_description:
|
||
en_US: used for searching
|
||
zh_Hans: 用于搜索网页内容
|
||
pt_BR: used for searching
|
||
llm_description: key words for searching
|
||
form: llm
|
||
- name: model
|
||
type: model-selector
|
||
scope: llm
|
||
required: true
|
||
label:
|
||
en_US: Model
|
||
zh_Hans: 使用的模型
|
||
pt_BR: Model
|
||
human_description:
|
||
en_US: Model
|
||
zh_Hans: 使用的模型
|
||
pt_BR: Model
|
||
llm_description: which Model to invoke
|
||
form: form
|
||
extra:
|
||
python:
|
||
source: tools/llm.py
|
||
```
|
||
|
||
请注意在该例子中指定了 `model` 的 `scope` 为 `llm`，那么此时用户就只能选择 `llm` 类型的参数，从而可以将上述用例的代码改成以下代码：
|
||
|
||
```python
|
||
from collections.abc import Generator
|
||
from typing import Any
|
||
|
||
from dify_plugin import Tool
|
||
from dify_plugin.entities.model.llm import LLMModelConfig
|
||
from dify_plugin.entities.tool import ToolInvokeMessage
|
||
from dify_plugin.entities.model.message import SystemPromptMessage, UserPromptMessage
|
||
|
||
class LLMTool(Tool):
|
||
def _invoke(self, tool_parameters: dict[str, Any]) -> Generator[ToolInvokeMessage]:
|
||
response = self.session.model.llm.invoke(
|
||
model_config=tool_parameters.get('model'),
|
||
prompt_messages=[
|
||
SystemPromptMessage(
|
||
content='you are a helpful assistant'
|
||
),
|
||
UserPromptMessage(
|
||
content=tool_parameters.get('query')
|
||
)
|
||
],
|
||
stream=True
|
||
)
|
||
|
||
for chunk in response:
|
||
if chunk.delta.message:
|
||
assert isinstance(chunk.delta.message.content, str)
|
||
yield self.create_text_message(text=chunk.delta.message.content)
|
||
```
|
||
|
||
### 调用 Summary
|
||
|
||
你可以请求该端点来总结一段文本，它会使用你当前 workspace 内的系统模型来总结文本。
|
||
|
||
**入口**
|
||
|
||
```python
|
||
self.session.model.summary
|
||
```
|
||
|
||
**端点**
|
||
|
||
* `text` 为需要被总结的文本。
|
||
* `instruction` 为你想要额外添加的指令，它可以让你风格化地总结文本。
|
||
|
||
```python
|
||
def invoke(
|
||
self, text: str, instruction: str,
|
||
) -> str:
|
||
```
|
||
|
||
### 调用 TextEmbedding
|
||
|
||
**入口**
|
||
|
||
```python
|
||
self.session.model.text_embedding
|
||
```
|
||
|
||
**端点**
|
||
|
||
```python
|
||
def invoke(
|
||
self, model_config: TextEmbeddingResult, texts: list[str]
|
||
) -> TextEmbeddingResult:
|
||
pass
|
||
```
|
||
|
||
### 调用 Rerank
|
||
|
||
**入口**
|
||
|
||
```python
|
||
self.session.model.rerank
|
||
```
|
||
|
||
**端点**
|
||
|
||
```python
|
||
def invoke(
|
||
self, model_config: RerankModelConfig, docs: list[str], query: str
|
||
) -> RerankResult:
|
||
pass
|
||
```
|
||
|
||
### 调用 TTS
|
||
|
||
**入口**
|
||
|
||
```python
|
||
self.session.model.tts
|
||
```
|
||
|
||
**端点**
|
||
|
||
```python
|
||
def invoke(
|
||
self, model_config: TTSModelConfig, content_text: str
|
||
) -> Generator[bytes, None, None]:
|
||
pass
|
||
```
|
||
|
||
请注意 `tts` 端点返回的 `bytes` 流是一个 `mp3` 音频字节流，每一轮迭代返回的都是一个完整的音频。如果你想做更深入的处理任务，请选择合适的库进行。
|
||
|
||
### 调用 Speech2Text
|
||
|
||
**入口**
|
||
|
||
```python
|
||
self.session.model.speech2text
|
||
```
|
||
|
||
**端点**
|
||
|
||
```python
|
||
def invoke(
|
||
self, model_config: Speech2TextModelConfig, file: IO[bytes]
|
||
) -> str:
|
||
pass
|
||
```
|
||
|
||
其中 `file` 是一个 `mp3` 格式编码的音频文件。
|
||
|
||
### 调用 Moderation
|
||
|
||
**入口**
|
||
|
||
```python
|
||
self.session.model.moderation
|
||
```
|
||
|
||
**端点**
|
||
|
||
```python
|
||
def invoke(self, model_config: ModerationModelConfig, text: str) -> bool:
|
||
pass
|
||
```
|
||
|
||
若该端点返回 `true` 则表示 `text` 中包含敏感内容。
|
||
|
||
## 相关资源
|
||
|
||
- [反向调用 Dify 服务](/plugin-dev-zh/9241-reverse-invocation) - 了解反向调用的根本概念
|
||
- [反向调用 App](/plugin-dev-zh/9242-reverse-invocation-app) - 了解如何调用平台内的 App
|
||
- [反向调用 Tool](/plugin-dev-zh/9242-reverse-invocation-tool) - 了解如何调用其它插件
|
||
- [模型插件开发指南](/plugin-dev-zh/0211-getting-started-new-model) - 学习如何开发自定义模型插件
|
||
- [模型设计规则](/plugin-dev-zh/0411-model-designing-rules) - 了解模型插件的设计原则
|
||
|
||
{/*
|
||
Contributing Section
|
||
DO NOT edit this section!
|
||
It will be automatically generated by the script.
|
||
*/}
|
||
|
||
---
|
||
|
||
[编辑此页面](https://github.com/langgenius/dify-docs/edit/main/plugin-dev-zh/9242-reverse-invocation-model.mdx) | [提交问题](https://github.com/langgenius/dify-docs/issues/new?template=docs.yml)
|
||
|