ChatGPTNextWeb
[Feature Request] Add Like/Dislike Feedback Buttons for RLHF Training
🥰 Feature Description
We are running an on-premise LLM and would like to further fine-tune the model using Reinforcement Learning from Human Feedback (RLHF). To facilitate this, we propose adding two buttons below each LLM-generated response: Like and Dislike.
When a user interacts with these buttons, the following data should be logged:
The user's feedback (like/dislike) The original prompt The LLM-generated response This data should be stored in either a log file or a database, allowing us to analyze user preferences and improve the model accordingly.
Use Case: This feature would help organizations running on-premise LLMs to collect structured user feedback, enabling them to refine their models for better performance and alignment with user expectations.
🧐 Proposed Solution
Add two UI buttons (👍 Like, 👎 Dislike) under each response. Store the feedback, prompt, and response in a structured format (log file or database). Provide configuration options to specify the storage method (file path or database connection). This feature would significantly enhance the usability of Next Chat for those looking to continuously improve their LLMs with human feedback.
Thank you for considering this request!
📝 Additional Information
No response
