GPT-Gradio-Agent

中文文档 | English

Build your own GPT Agent and manage your GPT-driven knowledge base with Azure OpenAI API (or OpenAI API) and Gradio!

Now I'm building a more user-friendly project called "RAGENT" and its plugin "RAGENT-Copilot", which not only implements the main features of the project, but also actually creates and calls Agents. If you want a better experience using, please check the repos.

Features

• 🤖 Chatbox Interface
• 🔧 detailed parameterization
• 📚️ GPT-driven(or any LLM-driven) knowledge base, and supports knowledge base management and RAG (Retrieval Enhanced Generation)
• 👓️ Excellent RAG retrieval capability (supports hybrid retrieval, reranking, specified file retrieval)
• 🖼️ Dall-E-3 Image Generator
• 🔍️ RAG Search
• 🖥️ Enables FULLY LOCAL embedding (Hugging Face) and chat (Ollama) (if you want OR don't have Azure OpenAI). This means that you can run GPT-Gradio-Agent's chat and knowledge base locally without connecting to the Azure OpenAI API!
• 📦 Packaged in a portable package, download and ready to use
• 🛜 Fetch web content and dialog via web scraper; and optionally store it in a knowledge base.

If you like this project, please star for it, this is the greatest encouragement to me!

Details

Basic Chatbox Interface

This is the basic chatbox interface where you can have a conversation directly with GPT, let them play the role of an expert and answer your questions with system prompts, and manage your multiple conversations.

GPT-driven knowledge base

In this interface, you can create and manage your own Knowledge Base (CRUD) and **have GPT answer your questions in conjunction with specified documents (or the entire Knowledge Base), enabling RAG (Retrieval Augmented Generation)

Very efficient!

Knowledge Base Management

Knowledge Base Details

v0.9.0 New: You can now view not only the file directory of the knowledge base, but also the specific content in the knowledge base, and understand the specific chunks.

Click the checkbox to view the specific information of the block:

It's important to note that when you load a knowledge base, it will automatically load out its specific information.

Also, it's a stand-alone module, which means you can use it to view Knowledge Base A when you open Knowledge Base B at the same time!

Local Embedding

v0.12.0 Added.

Now you can select Hugging Face(local) in Embedding Model Type and choose your own supported embedding model for vector embedding (currently supported: BGE v1.5 series).

Brand New Knowledge Base Management

Added in v0.12.0, this change is a leapfrog change and requires a manual migration of the knowledge base.

The knowledge base will now be stored centrally under the path . \knowledge base and is displayed as a drop-down list in the right sidebar. You can create a customized name for the knowledge base, which will be used as the name of the folder.

To migrate a legacy knowledge base, complete the following steps:

1. Create a new file embedding_config.json in the root directory and fill in the following contents:

{
    "your knowledge base name": {
        "embedding_model_type": "OpenAI",
        "embedding_model": "text-embedding-ada-002"
    },
    ...
}

Replace your knowledge base name with the name you want to use. 2. Move the original knowledge base folder in . \knowledge base and change the name of the file to the name entered in step 1. 3. Successful migration! Run the program to use it normally.

It's worth emphasizing that after selecting a knowledge base in the knowledge base list, the page is automatically updated with the corresponding embedding model information, but these information can be changed manually, my intention was that you could freely select the embedding model when creating the knowledge base, but this could actually result in two vectors of different dimensions in the same knowledge base (if you embedded it once and then take it upon yourself to modify the embedding model).

I don't have a good solution for this at the moment, so if you have a good idea, please feel free to submit an Issue. So for now, it is highly recommended that only when creating a new knowledge base adjust this option, and not modify it at any other time.

Dall-E-3 Image Generator

If you have access to the Azure OpenAI API, it's a waste not to try the Dall-E-3 model to generate images!

RAG Search

By harnessing the extensive knowledge of search engines (Bing Search) in conjunction with the exceptional capabilities of LLM, you can obtain precise answers and a wide range of references. It's inspired by search_with_lepton

You can also view previous history, which is expected to be available no later than v0.14.

Get Started

Git

0. Use git clone https://github.com/Wannabeasmartguy/GPT-Gradio-Agent.git to pull the codes；

Then open a Command Prompt (CMD) in the folder where the codes are stored, and use pip install -r requirements.txt to install the runtime environment.

1. Get Azure OpenAI API Key;

The model deployment name MUST be the same as the model name! For example, when deploying 'gpt-35-turbo', the deployment name of the model should also be 'gpt-35-turbo'. If you want to use Dall-E-3, you need to deploy it with the name 'dall-e-3'.

2. Copy .env_example and rename it to .env, change the environment variables: Starting from v0.9.0, there is a change in environmental variables: There is a conflict between Langchain's environment variable settings for OpenAI and Azure OpenAI, and it is not possible to use Azure Openai properly when both variables are set at the same time, which means that OpenAI cannot be supported at the same time from this version (I will look for a compatible method in the future), so you need to refer to the following environment variable settings. This also means that OpenAI cannot be supported from this release onwards (I will look for a way to make it compatible in the future), so you need to set the environment variables as follows.

AZURE_OAI_KEY: api key for Azure OpenAI; AZURE_OAI_ENDPOINT: Azure OpenAI's provided "endpoint"; API_VERSION: the version of the API used by Azure OpenAI; NOTE: use 2023-12-01-preview if you want to use Dall-E-3, 2023-09-15-preview and earlier will be deprecated on April 2, 2024; API_TYPE: indicates to use Azure OpenAI instead of OpenAI.

If you want to use RAG Search, set BING_SUBSCRIPTION_KEY and you can get access to related free resources in Azure with 1000 free calls per month.

3. Enjoy it!
   Use python GPT-Gradio-Agent.py in your terminal to run the the codes.You can see the URL in your terminal, and the default local URL is http://127.0.0.1:7860.

Release package

If you don't have Git installed, you can download the latest code from the release page.

After extracting, follow Step 1 and Step 2 above to configure the environment, and finally double-click Run.bat to run the code.

The default username is 'admin' and the password is '123456', which can be changed by yourself.The login function is turned off by default.

Todo List

• Adaptive language adaptation (Chinese, English supported)

• Use system-level prompts for role-playing

• Support for context quantity control

• Stream response

• Detailed configuration of additional parameters

• Choose models

  • Support gpt-4-turbo(which is called gpt-4-1106-preview in Azure)

  • Image input with gpt-4-turbo support (requires additional deployment of Azure's Computer Vision resources)

  • Generating images using Dall-E 3

• Dialogue management

• RAG(Retrieval Augmented Generation)

  • Multiple optimizations of searches -- ongoing

    • hybrid retrieval

    • reranking

    • specified file retrieval

  • Chat with single file

  • Summarize the file

  • Local knowledge base management

  • Chat with whole knowledge base

    • List citation sources

  • Estimated cost of embedding files

• RAG Web Search

  • Support Bing Search

  • Saving and Viewing Historical Search History

• Web scraper and interaction with knowledge base

• Import and export chat history

  • chat history export

  • chat history import

About