DiaryML/SETUP.md

DiaryML Setup Guide

Step-by-Step Installation

Step 1: Download Model Files

You need to download two GGUF model files. Go to: https://huggingface.co/huihui-ai/Huihui-Qwen3-VL-2B-Instruct-abliterated/tree/main/GGUF

Download these files:

• ggml-model-f16.gguf (~3.6 GB) - Main language model
• mmproj-model-f16.gguf (~300 MB) - Vision projection model

Alternative (if you have limited RAM/VRAM):

• huihui-qwen3-vl-2b-instruct-abliterated-q4_k_m.gguf (~1.5 GB) - Smaller quantized version

Place the downloaded files in the DiaryML/models/ folder.

Step 2: Install Python Dependencies

Open a terminal/command prompt in the DiaryML directory and run:

cd backend
pip install -r requirements.txt

This will install:

• FastAPI (web framework)
• SQLCipher (encrypted database)
• ChromaDB (vector database)
• llama-cpp-python (GGUF model runner)
• sentence-transformers (embeddings)
• transformers + torch (emotion detection)

Installation may take 5-10 minutes.

GPU Acceleration (Optional)

If you have an NVIDIA GPU and want faster AI inference:

# Uninstall CPU version
pip uninstall llama-cpp-python

# Install CUDA version
CMAKE_ARGS="-DLLAMA_CUBLAS=on" pip install llama-cpp-python

Step 3: Verify Installation

Run this to check everything is ready:

python -c "import fastapi, chromadb, transformers; print('All dependencies installed!')"

Step 4: Start DiaryML

On Windows:

start.bat

On Mac/Linux:

cd backend
python main.py

The server will start on http://localhost:8000

Step 5: First Launch

1. Open your browser to http://localhost:8000
2. You'll see the unlock screen
3. Enter a password (this creates your encrypted diary)
4. Wait for the AI model to load (~30 seconds)
5. Start journaling!

Troubleshooting

Error: "sqlcipher3" module not found

SQLCipher can be tricky on Windows. Try:

pip install pysqlcipher3

If that fails, you can temporarily disable encryption by modifying database.py:

# Comment out this line in get_connection():
# conn.execute(f"PRAGMA key = '{self.password}'")

Note: This removes encryption! Only use for testing.

Error: Model file not found

Make sure:

1. Files are named exactly ggml-model-f16.gguf and mmproj-model-f16.gguf
2. They are in the DiaryML/models/ folder
3. The files are fully downloaded (not partial)

Error: "CUDA out of memory"

The model is trying to use your GPU but running out of memory. Edit qwen_interface.py:

# Change this line (around line 37):
n_gpu_layers=-1,  # Change to 0 for CPU-only

Or use a smaller quantized model (q4_k_m).

Server won't start

Make sure port 8000 is not already in use:

# Windows
netstat -ano | findstr :8000

# Mac/Linux
lsof -i :8000

To use a different port, edit main.py at the bottom:

uvicorn.run(
    "main:app",
    host="127.0.0.1",
    port=8080,  # Change this
    reload=True
)

System Requirements

Minimum

• CPU: 4-core processor
• RAM: 4GB free
• Disk: 5GB (3GB models + 2GB for dependencies)
• OS: Windows 10+, macOS 10.15+, Linux

Recommended

• CPU: 8-core processor
• RAM: 8GB free
• GPU: NVIDIA GPU with 4GB+ VRAM (for faster inference)
• Disk: 10GB+

Performance Tips

1. Use GPU acceleration if available (5-10x faster)
2. Use quantized models (q4_k_m) if RAM is limited
3. Close other applications when using DiaryML
4. Reduce context window in qwen_interface.py if needed:

   n_ctx=2048,  # Instead of 4096
   

Next Steps

Once DiaryML is running:

1. Create your first journal entry
2. Try attaching an image
3. Chat with the AI about your entries
4. Check the mood timeline after a few entries
5. See daily suggestions each morning

Getting Help

If you encounter issues:

1. Check the console/terminal for error messages
2. Review the troubleshooting section above
3. Check the README.md for more details
4. Open an issue on GitHub with your error logs

---

Enjoy your private creative companion!