https://github.com/ysy1216/firewallm

By calling FirewaLLM, users can ensure the accuracy of the large model while greatly reducing the risk of privacy leakage when interacting with it. We believe that FirewallLLM is a privacy protected chatgpt interaction platform.
https://github.com/ysy1216/firewallm

chatbot chatgpt firewall flask llm nlp privacy python web

Last synced: 3 months ago
JSON representation

By calling FirewaLLM, users can ensure the accuracy of the large model while greatly reducing the risk of privacy leakage when interacting with it. We believe that FirewallLLM is a privacy protected chatgpt interaction platform.

• Host: GitHub
• URL: https://github.com/ysy1216/firewallm
• Owner: ysy1216
• Created: 2023-09-17T07:55:11.000Z (about 1 year ago)
• Default Branch: main
• Last Pushed: 2023-09-21T10:51:49.000Z (about 1 year ago)
• Last Synced: 2024-06-15T09:34:30.990Z (5 months ago)
• Topics: chatbot, chatgpt, firewall, flask, llm, nlp, privacy, python, web
• Language: Python
• Homepage:
• Size: 553 KB
• Stars: 39
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

• awesome-ChatGPT-repositories - FirewaLLM - By calling FirewaLLM, users can ensure the accuracy of the large model while greatly reducing the risk of privacy leakage when interacting with it. We believe that FirewallLLM is a privacy protected chatgpt interaction platform. (Chatbots)

README

        
# FirewaLLM

By calling FirewaLLM, users can ensure the accuracy of the large model while greatly reducing the risk of privacy leakage when interacting with it.

With the development of large models, more and more users are using them for interactive Q&A. However, with the increasing volume of data, privacy issues have attracted widespread attention.So, we launched a privacy protection small model. Firstly, we will judge sensitive statements based on user questions, and secondly, we will handle sensitive statements accordingly. We guarantee that user privacy exposure may be greatly reduced when interacting with the model. Finally, we ensure the accuracy of large model interactions by restoring sensitive information. Experiments have shown that users interact with large models by calling small models. 

The FirewaLLM framework not only protects user data privacy, but also protects the accuracy of the interaction model.

# Usage

1. Train FirewallM to have the ability to recognize sensitive information.

2. Open the FirewaLLM interface for interaction, and users can choose different large models for interaction.

```python

python FirewaLLM/app.py

```

2. Open the FirewaLLM interface for interaction, and users can choose different large models for interaction.

3. The effect is as follows:

Firstly. During the process of interacting with large models, FirewallLLM uses file filtering function to process sensitive information.

![image](https://github.com/ysy1216/FirewaLLM/blob/main/FirewaLLM_server.png) 

Secondly, FirewallLLM interacts data with user specified large models. Finally, the answer is processed to recover sensitive information, and the answer with the highest similarity to the question is returned.

![image](https://github.com/ysy1216/FirewaLLM/blob/main/FirewaLLM_fronted.png)

4. It can be intuitively seen that when a user inputs sensitive information, FirewaLLM will perform different levels of sensitive information processing operations on the sensitive information.

5. In the end, FirewaLLM will perform sensitivity restoration processing, returning the results with the highest similarity between the answers of the large model and the problem, in order to protect privacy while ensuring the accuracy of the overall process.