An Offline Mobile Conversational Agent for Mental Health Support: Learning from Emotional Dialogues and Psychological Texts with Student-Centered Evaluation
Overview
EmoSApp is an entirely offline conversational application designed to provide emotional and mental health support to students directly on your smartphone.
This app leverages the power of Large Language Models (LLMs), specifically a fine-tuned and quantized version of LLaMA-3.2-1B-Instruct.
All inference occurs locally on your device, ensuring complete privacy, accessibility, and data security.
The model was fine-tuned and quantized using the following frameworks:
- Executorch (v0.5.0) - https://github.com/pytorch/executorch
Used for lightweight, on-device deployment on Android smartphones.
- Torchtune (v0.6.0) - https://github.com/meta-pytorch/torchtune
Used for efficient fine-tuning and quantization of the LLaMA-3.2-1B-Instruct model.
Key Features
-
Fully Offline:
All inference happens on-device, with no internet connection required, making support available anytime, anywhere.
-
Enhanced Privacy:
All conversation logs and user data are stored locally on the user’s device, safeguarding sensitive mental health information.
-
Resource-Efficient:
Employs quantized model weights for smooth and responsive performance even on smartphones with as little as 6GB of system RAM.
-
Empathetic & Coherent:
Generates contextually relevant, understanding, and supportive responses to encourage emotional reflection and well-being.
Repository Contents
EmoSApp.apk: Android application package for EmoSApp installation on smartphones.Full_fine_tuned_model.pte: Contains the Full fine-tuned LLaMA-3.2-1B-Instruct model.QAT-LoRA_fine-tuned.pte: Contains the QAT-LoRA fine-tuned LLaMA-3.2-1B-Instruct model.tokenizer.bin: Tokenizer used by both model variants.Knowleadge_dataset.json: A curated dataset containing 14,582 question-answer (QA) pairs grouped into 1,459 conversational QA dialogues.
Setup and Installation for EmoSApp
-
Install the APK
- Download
EmoSApp.apk from this repository.
- Tap the APK file and install it on your Android smartphone.
-
Load Model & Tokenizer
- Open EmoSApp.
- Access the Settings menu (top-right bar).
- Tap “LOAD MODEL” to load both the model and tokenizer into your device memory.
- Wait for the confirmation message: “Successfully loaded model.”
-
Begin Conversation
- Once the model is successfully loaded, start chatting with our EmoSApp.
Knowledge Dataset
The knowledge dataset used for fine-tuning our model in this repository is derived exclusively from the following openly licensed, publicly available textbooks, each used in compliance with its respective Creative Commons (CC) license for non-commercial research:
- Neuroscience, Psychology and Conflict Management (2024) - Judith Rafferty, James Cook University, licensed under CC BY-NC 4.0 International.
- Fundamentals of Psychological Disorders (3rd Edition) (2024) - Alexis Bridley & Lee W. Daffin Jr., Washington State University, licensed under CC BY-NC-SA 4.0 International.
- Principles of Social Psychology (1st International H5P Edition) (2022) - Rajiv Jhangiani & Hammond Tarry, by BCcampus, licensed under CC BY-NC-SA 4.0 International.
- Essentials of Exercise and Sport Psychology: An Open Access Textbook (2021) - Zachary Zenko & Leighton Jones, Society for Transparency, Openness, and Replication in Kinesiology, licensed under CC BY 4.0 International.
- Psychology 2e (2014) - Rose M. Spielman, William J. Jenkins, Marilyn D. Lovett, licensed under CC BY 4.0 International.
- Essentials of Cognitive Psychology (2023) - Christopher Klein, University of North Alabama Digital Press, licensed under CC BY-NC-SA 4.0 International.
- Abnormal Psychology (2020) - Jorden A. Cummings, University of Saskatchewan Open Press, licensed under CC BY-NC-SA 4.0 International.
- Psychology: The Science of Human Potential (2019) - Jeffrey C. Levy, BCcampus, licensed under CC BY 4.0 International.
The dataset, app and associated models are distributed for academic, non-commercial research under the CC BY-NC-SA 4.0 International License.
Ethical Considerations
This repository and its contents are provided for non-clinical research purposes only. The application must not be used as a diagnostic, clinical advice or therapeutic tool.
- Individuals seeking help should always consult a licensed psychiatrist, psychologist, or qualified mental health professional.
- No warranties, express or implied, are provided regarding the accuracy, completeness, or reliability of model outputs.
- The authors and contributors shall not be held responsible for any errors, omissions, or outcomes resulting from the use of this app or dataset.
Contact
For any queries related to dataset, app and models please contact:
vimaleee2002@gmail.com
