Remi

Remi is a chatbot that utilizes a pre-trained language model, LL7M, to provide engaging conversations.

Getting started

Prerequisites

• Python 3.12 or higher
• torch and transformers libraries

Installation

1. Clone the repository. git clone https://github.com/alexlnkp/Remi.git
2. Create a Python virtual environment. python -m venv .venv
3. Install required libraries in the virtual environment. pip install -r infer.req

Usage

• Activate the virtual environment. source .venv/bin/activate (Linux) or . .venv/Scripts/activate (Windows)
• To run Remi in chatting mode, use python prompt.py or ./prompt.py (Linux only)

Contributing

Your contributions will be met with gratitude and will help greatly! If You would like to help Remi, please fork the repository and submit a pull request.

TODO

• General:

• Fine-tune the LLM to directly improve quality of the responses

• Do some fancy prompt-engineering.

• TTS related:

• Bootstrap a TTS in an external script to have an option to convert generated output directly to audio.

• ^ Using Bark, TTS can produce output with emotions. Therefore, also bootstrapping a tone recognition of the generated text would make the chatbot sound more natural.

• User input related:

• Bootstrap Whisper or any other voice recognition AI model to convert user speech into text which is then used as an input for inference of an LLM.

• Bootstrap the aforementioned tone recognition of the text to the user's voice recognition to add on layers of communications.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Credit

LL7M - the pre-trained LLM used by Remi.

Contact

For any questions or concerns, please contact the project maintainers at 413x1nkp@gmail.com.