Llama-3.1-8B-quantized-EEG-TimeLLM

EEG Forecasting with Llama 3.1-8B and Time-LLM

This repository contains the code and model for forecasting EEG signals by combining the quantized Llama 3.1-8B model from Hugging Face and a modified version of the Time-LLM framework.

Overview

This project aims to leverage large language models (LLMs) for time-series forecasting, specifically focusing on EEG data. The integration of Llama 3.1-8B allows us to apply powerful sequence modeling capabilities to predict future EEG signal patterns with high accuracy and efficiency.

Key Features

• Quantized Llama 3.1-8B Model: Utilizes a quantized version of Llama 3.1-8B to reduce computational requirements while maintaining performance.
• Modified Time-LLM Framework: Adapted the Time-LLM framework for EEG signal forecasting, allowing for efficient processing of EEG time-series data.
• Scalable and Flexible: The model can be easily adapted to other time-series forecasting tasks beyond EEG data.

Getting Started

Prerequisites

Before you begin, ensure you have the following installed:

• Python 3.8+
• PyTorch
• Transformers (Hugging Face)
• Time-LLM dependencies (see the original Time-LLM repository)
• Download the Llama 3.1-8B quantized model from Hugging Face:
  git lfs install
  git clone https://huggingface.co/akshathmangudi/llama3.1-8b-quantized

EEG datasets

The datasets can be get from this survey, choose the dataset you want to try.

Acknowledgments

• Hugging Face for the Llama 3.1-8B-quantized model.
• The original Time-LLM repository for the time-series framework.