git push origin main

Initial commit of mental_health_chatbot app

README.md

@@ -4,15 +4,24 @@ emoji: 🧠
 colorFrom: indigo
 colorTo: green
 sdk: streamlit
-sdk_version: 1.44.1
+sdk_version: 1.32.2
 app_file: app.py
 pinned: false
 license: mit
 ---
-
 # 🧠 Mental Health Counselor Assistant
 
-**Mental Health Counselor Assistant** is an AI-powered Streamlit application designed to help therapists and mental health professionals respond more effectively to patients. It combines machine learning for **intent classification** with powerful large language models (LLMs) like **Flan-T5**, **Mistral-7B**, and **Flan-Alpaca-GPT4** to generate emotionally intelligent, context-aware suggestions.
+**Mental Health Counselor Assistant** is an AI-powered Streamlit application hosted on [Hugging Face Spaces](https://huggingface.co/spaces/scdong/mental_health_chatbot). It assists mental health professionals by classifying user inputs and generating supportive counselor-style responses using large language models (LLMs) like **Flan-T5** and **Mistral-7B**.
+
+---
+
+## ⚡ Quick Start
+
+The app uses `google/flan-t5-base` for fast, low-latency response generation.
+
+> 📝 `app.py` is optimized for **speed**, using only `/flan-t5-base`.  
+> While the results may not be as expressive as other models, it loads and responds much faster.  
+> For higher quality but slower results, try the other two apps included.
 
 ---
 
@@ -20,154 +29,93 @@ license: mit
 
 ```
 mental_health_chatbot/
-├── app.py                                # Main app (fast Flan-T5 only)
-├── app_use_Mistral-7B.py                 # Local Mistral-7B (quantized .gguf)
-├── app_with_FlanT5_FlanAlpacaGPT4_FlanUL2.py  # App with selectable hosted Hugging Face LLMs
-├── requirements.txt                      # Python dependencies
-├── Dockerfile                            # For containerized deployment
-├── README.md                             # You're here!
-├── LICENSE                               # Open-source license
+├── app.py                                # Fast demo with Flan-T5 only (used on Hugging Face)
+├── app_use_Mistral-7B.py                 # Local LLM with quantized Mistral-7B via llama.cpp
+├── app_with_FlanT5_FlanAlpacaGPT4_FlanUL2.py  # Hosted Hugging Face models (slower, richer output)
+├── requirements.txt
+├── Dockerfile
+├── README.md
+├── LICENSE
 
 ├── dataset/
-│   └── Kaggle_Mental_Health_Conversations_train.csv  # Training dataset
+│   └── Kaggle_Mental_Health_Conversations_train.csv
 
 ├── utils/
-│   └── helper_functions.py               # Preprocessing and helpers
+│   └── helper_functions.py
 
 ├── notebooks/
-│   ├── Flan-T5_on_Kaggle_Dataset.ipynb       # Inference experiments
-│   ├── Mistral-7B_on_Kaggle_Dataset.ipynb    # Local LLM generation
-│   └── ML_on_Kaggle_Dataset.ipynb            # Model training and eval
+│   ├── Flan-T5_on_Kaggle_Dataset.ipynb
+│   ├── Mistral-7B_on_Kaggle_Dataset.ipynb
+│   └── ML_on_Kaggle_Dataset.ipynb
 
 └── log/
-    └── chat_log_2025-04-11_05-31-25.csv      # Example saved conversation log
+    └── chat_log_2025-04-11_05-31-25.csv
 ```
 
 ---
 
 ## 💡 Features
 
-- 🔍 **Response Type Classification**  
-  Predicts one of four counseling intents from patient input:
-  - `advice`
-  - `validation`
-  - `information`
-  - `question`
+- 🔍 **Intent Classification** (XGBoost):  
+  Tags inputs as `advice`, `validation`, `information`, or `question`.
 
-- 🧠 **LLM-Powered Reply Generation**  
-  Generates suggestions using:
-  - ✅ `google/flan-t5-base` (fast, hosted)
-  - ✅ `mistralai/Mistral-7B-Instruct-v0.1` (runs locally via `llama.cpp`)
-  - ✅ `declare-lab/flan-alpaca-gpt4-xl` and `flan-ul2` (hosted via Hugging Face)
+- 🤖 **LLM-Based Suggestion Generation**  
+  Choose from:
+  - `google/flan-t5-base` (fastest)
+  - `declare-lab/flan-alpaca-gpt4-xl`
+  - `google/flan-ul2`
+  - `mistralai/Mistral-7B-Instruct-v0.1` (local only)
+  
 
-⚠️ **Note**: LLM response times may vary. Hosted models are slower but provide higher-quality, human-like responses.
-
-- 🎯 **Primed & Prompt-Engineered Outputs**  
-  Uses few-shot prompt design to ensure counselor-style tone, empathy, and helpfulness.
-
-- 🧪 **Modular & Experiment-Ready**  
-  Jupyter notebooks for training, testing, and benchmarking models.
-
-- 💾 **Conversation Logging**  
-  Auto-saves counselor-patient interactions to `/log` for post-session review.
+- 💾 **Session Logging**  
+  Automatically saves conversation history to CSV in `/log`.
 
 ---
 
-## 🚀 Getting Started
-
-### 1. Clone the Repository
+## 🚀 Run Locally (Optional)
 
 ```bash
-git clone https://github.com/your-username/mental_health_chatbot.git
+git clone https://huggingface.co/spaces/scdong/mental_health_chatbot
 cd mental_health_chatbot
-```
-
-### 2. Set Up the Environment
 
-```bash
 python -m venv venv
-source venv/bin/activate  # Windows: venv\Scripts\activate
+source venv/bin/activate
 pip install -r requirements.txt
-```
-
-### 3. Run the App
 
-#### Run fast, default version (Flan-T5):
-
-```bash
 streamlit run app.py
 ```
 
-#### Run version with all Hugging Face LLMs:
-
-```bash
-streamlit run app_with_FlanT5_FlanAlpacaGPT4_FlanUL2.py
-```
-
-#### Run local Mistral-7B (with `llama-cpp-python`):
+For the Mistral or full-model version:
 
 ```bash
 streamlit run app_use_Mistral-7B.py
-```
-
----
-
-## 🐳 Docker (Optional)
-
-To build and deploy with Docker:
-
-```bash
-docker build -t counselor-assistant .
-docker run -p 8501:8501 counselor-assistant
+# or
+streamlit run app_with_FlanT5_FlanAlpacaGPT4_FlanUL2.py
 ```
 
 ---
 
 ## 📓 Dataset
 
-This project uses the Kaggle dataset:  
-**Kaggle_Mental_Health_Conversations_train.csv**  
-Located in the `dataset/` folder, it contains real user mental health exchanges labeled with response types.
-
----
-
-## 📔 Notebooks
-
-Explore or reproduce training and benchmarking in:
-
-- `ML_on_Kaggle_Dataset.ipynb`: Training XGBoost and TF-IDF classifier
-- `Flan-T5_on_Kaggle_Dataset.ipynb`: Hosted inference test
-- `Mistral-7B_on_Kaggle_Dataset.ipynb`: Local quantized LLM
-
----
-
-## 📂 Log Files
-
-Each chat session is saved with timestamp in the `log/` directory, e.g.:
-
-```
-log/chat_log_2025-04-11_05-31-25.csv
-```
-
-These CSVs include patient messages, predicted intent, AI-generated counselor suggestions, and classification confidence.
+This project uses:
+- 🧾 `Kaggle_Mental_Health_Conversations_train.csv`  
+  A labeled dataset of user statements and counselor responses used for training and testing classification.
 
 ---
 
 ## 🧑‍⚕️ Intended Use
 
-This tool is built for **prototyping** and **research**. Potential users include:
-
-- Mental health professionals exploring AI support
-- Researchers in NLP and digital health
-- Developers building therapeutic chatbot systems
+This tool is designed for:
+- Mental health professionals exploring AI assistance
+- NLP researchers and students
+- Developers building mental health chatbot prototypes
 
-⚠️ This tool is **not a replacement for human therapy**.
+> ⚠️ **Not a substitute for professional mental health advice.**
 
 ---
 
 ## 📜 License
 
-Licensed under the **MIT License**.  
-See [`LICENSE`](./LICENSE) for full terms.
-
----
+This project is licensed under the **MIT License**.  
+See the `LICENSE` file for more details.
+```