The AllyArc project leverages conversational AI to create an engaging and supportive educational experience for autistic learners. This repository contains the development lifecycle of the AllyArc chatbot, from dataset creation, through model fine-tuning, to performance evaluation. Initially, the project explored the use of the LLaMA model, but due to limitations in conversational capabilities, it pivoted to the GPT-3.5 Turbo model.

To replicate or build upon this project, you will need:

• Python 3.8 or later
• Access to Hugging Face and OpenAI APIs
• Jupyter Notebook or JupyterLab

First, clone this repository to your local machine:

git clone https://github.com/yourgithubusername/allyarc-project.git
cd allyarc-project

Install the required Python packages:

pip install -r requirements.txt

Initial dataset construction for the chatbot.

Ensure you have Jupyter Notebook installed and run:

jupyter notebook AllyArc_Dataset_Chat.ipynb

This notebook outlines the steps to compile and preprocess the initial dataset. Follow the instructions within to generate your dataset.

Dataset uploaded to Hugging Face: AllyArc/chat_dataset.

Fine-tuning the LLaMA model with the prepared dataset.

Launch this notebook in Jupyter:

jupyter notebook Share_FineTuneAllyArc_llama.ipynb

Details the process of fine-tuning the LLaMA model. Execute each cell as per the instructions provided within.

Fine-tuned model on Hugging Face: AllyArc/llama_allyarc.

Reformatting the dataset for GPT-3.5 Turbo model fine-tuning.

Run the notebook via Jupyter:

jupyter notebook AllyArc_Dataset_OAI_Format.ipynb

Adjusts the dataset structure for OpenAI model compatibility. Follow the notebook steps to modify and export your dataset.

Reformatted dataset on Hugging Face: AllyArc/allyarc_oai_format.

Validate the dataset format for OpenAI fine-tuning.

Execute the notebook in Jupyter:

jupyter notebook check_format.ipynb

Ensures the dataset is correctly formatted for fine-tuning. Execute the cells and follow any corrective actions suggested.

Evaluate the fine-tuned GPT-3.5 Turbo model's performance.

Launch with Jupyter:

jupyter notebook eval_oai.ipynb

This notebook assesses the fine-tuned model using specific prompts. Follow the notebook to run the evaluation and view the results.

We welcome contributions to the AllyArc project. Please open an issue to discuss your ideas or submit pull requests for consideration.

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