DeepSeek Local

A local, private interface for running DeepSeek language models on your own machine. All processing happens locally - files and queries never leave your computer.

Demo

Features

• Run DeepSeek language models entirely on your local machine
• Upload and process files privately
• Optimized for systems with 8GB GPU memory using 4-bit quantization
• Support for various DeepSeek models including DeepSeek-Coder
• Code-specific formatting with syntax highlighting

Installation

Suggestion: use pip freeze to ensure the floor doesn't become lava later.

Option 1: Using the Installation Script (Recommended)

1. Clone this repository:

   git clone https://github.com/cschladetsch/deepseek-local.git
   cd deepseek-local

2. Run the installation script:

   chmod +x install_deepseek.sh
   ./install_deepseek.sh

3. During installation, you'll be prompted for your Hugging Face token if you want to use gated models.

4. Install additional dependencies for code formatting:

   source venv/bin/activate
   pip install markdown pygments

Option 2: Manual Installation

1. Clone this repository:

   git clone https://github.com/yourusername/deepseek-local.git
   cd deepseek-local

2. Create and activate a virtual environment:

   python3 -m venv venv
   source venv/bin/activate

3. Install the required dependencies:

   pip install -r requirements.txt

4. Download a model from Hugging Face:

   mkdir -p models
   cd models
   git clone https://huggingface.co/deepseek-ai/deepseek-coder-6.7b-instruct
   cd ..

Usage

Starting the Interface

1. Start the local interface:

   ./start_deepseek.sh

2. To share with other devices on your network:

   ./start_deepseek_network.sh

3. Access the interface at http://127.0.0.1:7860 in your web browser

Command-line Options

The installation script supports various options:

./install_deepseek.sh [options]

Options:
  -h, --help                  Show help message
  -m, --model MODEL_ID        Specify model ID (e.g., deepseek-ai/deepseek-v2)
  -l, --list                  List available recommended models
  -s, --small                 Use smaller models (for systems with less RAM)
  --no-auth                   Skip Hugging Face authentication
  --cleanup                   Remove temporary files and fix permissions
  --uninstall                 Remove the installation completely

The startup script also supports options:

./start_deepseek.sh [options]

Options:
  -p, --port PORT    Specify port (default: 7860)
  -m, --model DIR    Specify model directory name (default: auto-detect)
  -h, --help         Show help

Supported Models

Models are downloaded from Hugging Face. Some recommended models:

Large models (7B+ parameters, 32GB+ RAM recommended)

• deepseek-ai/deepseek-v2 (requires auth)
• deepseek-ai/deepseek-coder-33b-instruct (requires auth)
• mistralai/Mistral-7B-Instruct-v0.2

Medium models (3-7B parameters, 16GB+ RAM recommended)

• deepseek-ai/deepseek-coder-6.7b-instruct
• microsoft/phi-3-mini-4k-instruct

Small models (1-3B parameters, 4GB+ RAM)

• deepseek-ai/deepseek-coder-1.3b-instruct
• microsoft/phi-2

System Requirements

• Ubuntu 20.04+ or Windows WSL2
• Python 3.8+
• 8GB RAM minimum (16GB+ recommended for medium models)
• NVIDIA GPU with 8GB VRAM (for GPU acceleration) or CPU-only mode

Customization

You can customize the interface by editing:

• style.css - For UI appearance
• deepseek_repl.py - For functionality changes

Project Structure

deepseek-local/
- install_deepseek.sh       # Installation script
- start_deepseek.sh         # Local startup script
- start_deepseek_network.sh # Network-accessible startup script
- deepseek_repl.py          # Main Python interface
- style.css                 # CSS styling
- requirements.txt          # Python dependencies
- models/                   # Downloaded models
�   - deepseek-coder-6.7b-instruct/  # Example model
- venv/                     # Python virtual environment

Dependencies

The project requires several Python packages listed in requirements.txt. The main dependencies are:

• torch - For deep learning operations
• transformers - For loading and running the models
• gradio - For the web interface
• bitsandbytes - For model quantization
• markdown and pygments - For code formatting

You can install all dependencies using:

pip install -r requirements.txt

Privacy

All processing happens locally on your machine. Files uploaded to the interface and all queries are processed entirely on your local hardware and are never sent to external servers.

Troubleshooting

Port Already in Use

If port 7860 is already in use, specify a different port:

./start_deepseek.sh --port 7861

Out of Memory Errors

If you encounter GPU memory errors, try a smaller model:

./install_deepseek.sh --model deepseek-ai/deepseek-coder-1.3b-instruct

Permission Issues

If you encounter permission issues:

./install_deepseek.sh --cleanup

Missing Dependencies

If you see errors about missing Python modules:

source venv/bin/activate
pip install -r requirements.txt

License

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