Skip to content

赵月辉 / fastgpt-migrated

Go to a project
Commit b6f9f77e by Archer Committed by GitHub
Options

chat model config (#158) 

* chat model config

* fix: i18n next
parent da4b14fb
Showing with 232 additions and 33 deletions
client/next.config.js
...@@ -21,13 +21,17 @@ const nextConfig = { ...@@ -21,13 +21,17 @@ const nextConfig = {
asyncWebAssembly: true, asyncWebAssembly: true,
layers: true layers: true
}; };
config.module.rules = config.module.rules.concat([ config.module = {
{ ...config.module,
test: /\.svg$/i, rules: config.module.rules.concat([
issuer: /\.[jt]sx?$/, {
use: ['@svgr/webpack'] test: /\.svg$/i,
} issuer: /\.[jt]sx?$/,
]); use: ['@svgr/webpack']
}
]),
exprContextCritical: false
};
return config; return config;
} }
......
client/src/constants/flow/ModuleTemplate.ts
...@@ -261,7 +261,8 @@ export const KBSearchModule: FlowModuleTemplateType = { ...@@ -261,7 +261,8 @@ export const KBSearchModule: FlowModuleTemplateType = {
{ {
key: 'quoteQA', key: 'quoteQA',
label: '引用内容', label: '引用内容',
description: '搜索结果为空时不返回', description:
'始终返回数组，如果希望搜索结果为空时执行额外操作，需要用到上面的两个输入以及目标模块的触发器',
type: FlowOutputItemTypeEnum.source, type: FlowOutputItemTypeEnum.source,
valueType: FlowValueTypeEnum.kbQuote, valueType: FlowValueTypeEnum.kbQuote,
targets: [] targets: []
...@@ -916,7 +917,8 @@ export const appTemplates: (AppItemType & { avatar: string; intro: string })[] = ...@@ -916,7 +917,8 @@ export const appTemplates: (AppItemType & { avatar: string; intro: string })[] =
{ {
key: 'quoteQA', key: 'quoteQA',
label: '引用内容', label: '引用内容',
description: '搜索结果为空时不返回', description:
'始终返回数组，如果希望搜索结果为空时执行额外操作，需要用到上面的两个输入以及目标模块的触发器',
type: 'source', type: 'source',
valueType: 'kb_quote', valueType: 'kb_quote',
targets: [ targets: [
...@@ -1892,7 +1894,8 @@ export const appTemplates: (AppItemType & { avatar: string; intro: string })[] = ...@@ -1892,7 +1894,8 @@ export const appTemplates: (AppItemType & { avatar: string; intro: string })[] =
{ {
key: 'quoteQA', key: 'quoteQA',
label: '引用内容', label: '引用内容',
description: '搜索结果为空时不返回', description:
'始终返回数组，如果希望搜索结果为空时执行额外操作，需要用到上面的两个输入以及目标模块的触发器',
type: 'source', type: 'source',
valueType: 'kb_quote', valueType: 'kb_quote',
targets: [ targets: [
......
docSite/docs/develop/data_config/chat_models.md 0 → 100644
---
sidebar_position: 2
---
# Other Chat Model Configuration
By default, FastGPT is only configured with 3 models of GPT. If you need to integrate other models, you need to do some additional configuration.
## 1. Install OneAPI
First, you need to deploy a [OneAPI](/docs/develop/oneapi) and add the corresponding "channel".
![](./imgs/chatmodels1.png)
## 2. Add FastGPT Configuration
You can find the configuration file in /client/src/data/config.json (for local development, you need to copy it as config.local.json). In the configuration file, there is a section for chat model configuration:
```json
"ChatModels": [
{
"model": "gpt-3.5-turbo", // The model here needs to correspond to the model in OneAPI
"name": "FastAI-4k", // The name displayed externally
"contextMaxToken": 4000, // Maximum context token, calculated according to GPT35 regardless of the model. Models other than GPT need to roughly calculate this value themselves. You can call the official API to compare the token ratio and then roughly calculate it here.
// For example: the ratio of Chinese and English tokens in Wenxin Yiyuan is basically 1:1, while the ratio of Chinese tokens in GPT is 2:1. If the maximum token of Wenxin Yiyuan is 4000, then you can fill in 8000 here, or fill in 7000 for safety.
"quoteMaxToken": 2000, // Maximum token for quoting knowledge base
"maxTemperature": 1.2, // Maximum temperature
"price": 1.5, // Price per token => 1.5 / 100000 * 1000 = 0.015 yuan/1k token
"defaultSystem": "" // Default system prompt
},
{
"model": "gpt-3.5-turbo-16k",
"name": "FastAI-16k",
"contextMaxToken": 16000,
"quoteMaxToken": 8000,
"maxTemperature": 1.2,
"price": 3,
"defaultSystem": ""
},
{
"model": "gpt-4",
"name": "FastAI-Plus",
"contextMaxToken": 8000,
"quoteMaxToken": 4000,
"maxTemperature": 1.2,
"price": 45,
"defaultSystem": ""
}
],
```
### Add a New Model
Taking Wenxin Yiyuan as an example:
```json
"ChatModels": [
...
{
"model": "ERNIE-Bot",
"name": "Wenxin Yiyuan",
"contextMaxToken": 4000,
"quoteMaxToken": 2000,
"maxTemperature": 1,
"price": 1.2
}
...
]
```
After adding it, restart the application and you can choose the Wenxin Yiyuan model for conversation.
docSite/docs/develop/data_config/intro.md 0 → 100644
---
sidebar_position: 1
---
# Quick Introduction
Due to the limitations of environment variables in configuring complex content, the new version of FastGPT uses ConfigMap to mount configuration files. You can find the default configuration file at client/data/config.json.
In the development environment, you need to make a copy of config.json as config.local.json for it to take effect.
This configuration file includes customization of the frontend page, system-level parameters, and AI dialogue models, etc.
## Brief Explanation of Basic Fields
Here, we will introduce some basic configuration fields.
```json
// This configuration controls some styles of the frontend
"FeConfig": {
"show_emptyChat": true, // Whether to display the introduction page when the conversation page is empty
"show_register": false, // Whether to display the registration button (including forget password, register account, and third-party login)
"show_appStore": false, // Whether to display the app store (currently the permissions are not properly set, so it is useless to open it)
"show_userDetail": false, // Whether to display user details (account balance, OpenAI binding)
"show_git": true, // Whether to display Git
"systemTitle": "FastAI", // The title of the system
"authorText": "Made by FastAI Team.", // Signature
"gitLoginKey": "" // Git login credentials
}
```
```json
// This configuration file contains system-level parameters
"SystemParams": {
"gitLoginSecret": "", // Git login credentials
"vectorMaxProcess": 15, // Maximum number of processes for vector generation, set in combination with database performance and key
"qaMaxProcess": 15, // Maximum number of processes for QA generation, set in combination with database performance and key
"pgIvfflatProbe": 20 // pg vector search probe. Can be ignored before setting up the index, usually only needed for more than 500,000 groups.
},
```
docSite/docs/intro.md
...@@ -2,47 +2,45 @@ ...@@ -2,47 +2,45 @@
sidebar_position: 1 sidebar_position: 1
--- ---
# 快速了解 FastGpt # Introduction to FastGpt
FastGPT 是一个基于 LLM 大语言模型的知识库问答系统，提供开箱即用的数据处理、模型调用等能力。同时可以通过 Flow 可视化进行工作流编排，从而实现复杂的问答场景！ FastGPT is a knowledge-based question-answering system built on top of the LLM language model. It provides out-of-the-box capabilities for data processing, model invocation, and more. Additionally, it offers a visual workflow editor called Flow to enable complex question-answering scenarios!
| | |
| | |
| -------------------------- | -------------------------- | | -------------------------- | -------------------------- |
| ![Demo](./imgs/intro1.png) | ![Demo](./imgs/intro2.png) | | ![Demo](./imgs/intro1.png) | ![Demo](./imgs/intro2.png) |
| ![Demo](./imgs/intro3.png) | ![Demo](./imgs/intro4.png) | | ![Demo](./imgs/intro3.png) | ![Demo](./imgs/intro4.png) |
## FastGPT 能力 ## FastGPT Capabilities
### 1. AI 客服 ### 1. AI Customer Service
通过导入文档或已有问答对进行训练，让 AI 模型能根据你的文档进行回答问题。 Train the AI model by importing documents or existing question-answer pairs, allowing it to answer questions based on your documents.
![Ability1](./imgs/ability1.png) ![Ability1](./imgs/ability1.png)
### 2. 自动数据预处理 ### 2. Automated Data Preprocessing
提供手动输入、直接分段、LLM 自动处理和 CSV 等多种数据导入途径，兼顾精确和快速训练场景。 Provides multiple ways to import data, including manual input, direct segmentation, LLM automatic processing, and CSV, to accommodate precise and fast training scenarios.
![Ability1](./imgs/ability2.png) ![Ability1](./imgs/ability2.png)
### 3. 工作流编排 ### 3. Workflow Orchestration
基于 Flow 模块的工作流编排，可以帮助你设计更加复杂的问答流程。例如查询数据库、查询库存、预约实验室等。
Design more complex question-answering workflows using the Flow module. For example, querying databases, checking inventory, scheduling laboratory appointments, etc.
![Ability1](./imgs/ability3.png) ![Ability1](./imgs/ability3.png)
### 4. 无缝衔接的 OpenAPI ### 4. Seamless Integration with OpenAPI
FastGPT 对外 API 接口对齐 GPT 官方接口，你可以直接在现有的 GPT 应用中通过修改 BaseURL 和 Authorization 即可接入 FastGPT。 FastGPT's API interface aligns with the official GPT API, allowing you to integrate FastGPT into existing GPT applications by simply modifying the BaseURL and Authorization.
![Ability1](./imgs/ability4.png) ![Ability1](./imgs/ability4.png)
## FastGPT 特点 ## FastGPT Features
1. 项目开源。FastGPT 遵循 Apache License 2.0 开源协议，你可以在 GitHub Clone FastGPT 进行二次开发和发布。FastGPT 社区版将保留核心的功能，商业版仅在社区版基础上使用 API 的形式进行扩展，不影响学习使用。 1. Open-source project. FastGPT follows the Apache License 2.0 open-source agreement. You can clone FastGPT from GitHub for further development and distribution. The community edition of FastGPT will retain core functionality, while the commercial edition extends it through APIs without affecting learning usage.
2. 独特的 QA 结构。针对客服问答场景设计的 QA 结构，提高在大量数据场景中的准确性。 2. Unique QA structure. Designed specifically for customer service question-answering scenarios, the QA structure improves accuracy in large-scale data scenarios.
3. 可视化工作流。通过 Flow 模块展示了从问题输入到模型输出的完整流程，便于调试和设计复杂流程。 3. Visual workflow. The Flow module displays the complete process from question input to model output, facilitating debugging and designing complex workflows.
4. 无限扩展。基于 HTTP 进行扩展，无需修改 FastGPT 源码，也可快速接入现有的程序中。 4. Infinite scalability. Extensible via HTTP without modifying FastGPT source code, allowing for quick integration into existing programs.
5. 便捷调试。提供搜索测试、引用修改、完整对话预览等多种调试途径。 5. Convenient debugging. Provides various debugging methods, such as search testing, reference modification, and complete conversation preview.
6. 支持多种模型：支持 GPT、Claude、文心一言等多类 LLM 模型，未来也将支持自定义的向量模型。 6. Support for multiple models: Supports various LLM models such as GPT, Claude, and Wenxin Yiyuan, and will also support custom vector models in the future.
## 知识库核心流程图 ## Core Knowledge Base Process Diagram
![KBProcess](./imgs/KBProcess.jpg?raw=true 'KBProcess') ![KBProcess](./imgs/KBProcess.jpg?raw=true 'KBProcess')
docSite/i18n/zh-Hans/docusaurus-plugin-content-docs/current.json
...@@ -8,6 +8,9 @@ ...@@ -8,6 +8,9 @@
"sidebar.docSidebar.category.Proxy": { "sidebar.docSidebar.category.Proxy": {
"message": "Proxy 方案" "message": "Proxy 方案"
}, },
"sidebar.docSidebar.category.Data Config": {
"message": "Config 配置"
},
"sidebar.docSidebar.category.Deploy": { "sidebar.docSidebar.category.Deploy": {
"message": "部署" "message": "部署"
}, },
......
docSite/i18n/zh-Hans/docusaurus-plugin-content-docs/current/develop/data_config/chat_models.md 0 → 100644
---
sidebar_position: 2
---
# 其他对话模型配置
默认情况下，FastGPT 只配置了 GPT 的 3 个模型，如果你需要接入其他模型，需要进行一些额外配置。
## 一、安装 OneAPI
首先你需要部署一个 [OneAPI](/docs/develop/oneapi)，并添加对应的【渠道】
![](./imgs/chatmodels1.png)
## 二、添加 FastGPT 配置
可以在 /client/src/data/config.json 里找到配置文件（本地开发需要复制成 config.local.json），配置文件中有一项是对话模型配置：
```json
"ChatModels": [
{
"model": "gpt-3.5-turbo", // 这里的模型需要对应 OneAPI 的模型
"name": "FastAI-4k", // 对外展示的名称
"contextMaxToken": 4000, // 最大长下文 token，无论什么模型都按 GPT35 的计算。GPT 外的模型需要自行大致计算下这个值。可以调用官方接口去比对 Token 的倍率，然后在这里粗略计算。
// 例如：文心一言的中英文 token 基本是 1:1，而 GPT 的中文 Token 是 2:1，如果文心一言官方最大 Token 是 4000，那么这里就可以填 8000，保险点就填 7000.
"quoteMaxToken": 2000, // 引用知识库的最大 Token
"maxTemperature": 1.2, // 最大温度
"price": 1.5, // 1个token 价格 => 1.5 / 100000 * 1000 = 0.015元/1k token
"defaultSystem": "" // 默认的系统提示词
},
{
"model": "gpt-3.5-turbo-16k",
"name": "FastAI-16k",
"contextMaxToken": 16000,
"quoteMaxToken": 8000,
"maxTemperature": 1.2,
"price": 3,
"defaultSystem": ""
},
{
"model": "gpt-4",
"name": "FastAI-Plus",
"contextMaxToken": 8000,
"quoteMaxToken": 4000,
"maxTemperature": 1.2,
"price": 45,
"defaultSystem": ""
}
],
```
### 添加新模型
以添加文心一言为例:
```json
"ChatModels": [
...
{
"model": "ERNIE-Bot",
"name": "文心一言",
"contextMaxToken": 4000,
"quoteMaxToken": 2000,
"maxTemperature": 1,
"price": 1.2
}
...
]
```
添加完后，重启应用即可在选择文心一言模型进行对话。
docSite/docs/develop/dataconfig/intro.md → docSite/i18n/zh-Hans/docusaurus-plugin-content-docs/current/develop/data_config/intro.md
...@@ -2,7 +2,7 @@ ...@@ -2,7 +2,7 @@
sidebar_position: 1 sidebar_position: 1
--- ---
# Config 配置文件说明 # 快速介绍
由于环境变量不利于配置复杂的内容，新版 FastGPT 采用了 ConfigMap 的形式挂载配置文件，你可以在 client/data/config.json 看到默认的配置文件。 由于环境变量不利于配置复杂的内容，新版 FastGPT 采用了 ConfigMap 的形式挂载配置文件，你可以在 client/data/config.json 看到默认的配置文件。
......
docSite/sidebars.js
...@@ -28,6 +28,19 @@ const sidebars = { ...@@ -28,6 +28,19 @@ const sidebars = {
'develop/dev', 'develop/dev',
{ {
type: 'category', type: 'category',
label: 'Data Config',
link: {
type: 'generated-index'
},
items: [
{
type: 'autogenerated',
dirName: 'develop/data_config'
}
]
},
{
type: 'category',
label: 'Deploy', label: 'Deploy',
collapsed: false, collapsed: false,
link: { link: {
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or sign in to comment