Zord Coder v1

Zord Coder v1 - The ultimate multilingual coding assistant optimized for Android Termux. Built with model merging techniques, nanoGPT philosophy, and blazing-fast GGUF inference.

Features

• Blazing Fast - Optimized GGUF quantization for mobile devices
• Multi-Language - Python, JavaScript, TypeScript, C++, Rust, Go, Java, Bash, and more
• Interactive CLI - Beautiful terminal interface with syntax highlighting
• Streaming Output - Real-time token generation
• Reasoning Mode - Chain-of-thought like DeepSeek-R1
• Termux Ready - Optimized for Android devices

Quick Start

1. Clone & Run

git clone https://github.com/sajadkoder/zordcoder.git
cd zordcoder
pip install -r requirements.txt
python scripts/zord_cli.py --interactive

The model will download automatically on first run!

Note: First run will download the model (~833MB) automatically from HuggingFace.

Installation (Termux)

# Run the setup script
bash scripts/setup_termux.sh

# Or manual installation
pkg update && pkg install python git
pip install -r requirements.txt

# Download model and run
python3 scripts/zord_cli.py --interactive

Usage

Interactive Mode

python3 scripts/zord_cli.py --interactive

Single Prompt

python3 scripts/zord_cli.py "Write a Python hello world"

With Custom Settings

python3 scripts/zord_cli.py "Explain recursion" \
    --temperature 0.2 \
    --max-tokens 512 \
    --context 2048

Commands

Command	Description
clear	Clear conversation history
reasoning	Toggle reasoning mode
stream	Toggle streaming mode
metrics	Show performance metrics
history	Show conversation history
language <lang>	Set preferred language
help	Show help message
exit	Exit the program

Project Structure

zordcoder/
├── config/
│   └── merge_config.yaml      # Model merging configuration
├── docs/
│   ├── MODEL_SELECTION.md    # Model selection strategy
│   ├── CONVERSION.md         # GGUF conversion guide
│   └── OPTIMIZATION.md       # Performance optimization
├── scripts/
│   ├── setup_termux.sh       # Termux installation script
│   ├── train_zord.py         # nanoGPT training script
│   └── zord_cli.py           # CLI interface
├── src/
│   └── zord_core.py          # Core inference engine
├── .gitignore
├── README.md
└── requirements.txt

Documentation

• Model Selection Strategy
• GGUF Conversion Guide
• Performance Optimization

System Requirements

Minimum

• Android 7.0+
• 3GB RAM
• 2GB Storage

Recommended

• Android 10+
• 6GB+ RAM
• 4GB+ Storage

Model Merging

Zord Coder v1 uses state-of-the-art model merging techniques:

• TIES (Task Interference Elimination)
• DARE (Data-Aware Reward Estimation)
• SLERP interpolation

See MODEL_SELECTION.md for details.

Performance

Metric	Value
Tokens/sec	15-30 (Q4_K_M)
Cold Start	~3 seconds
Memory Usage	~2GB (Q4_K_M)
Context Length	2048 tokens

Troubleshooting

Model Not Loading

# Check model file
ls -la models/

# Verify file
file models/zordcoder-v1-q4_k_m.gguf

Out of Memory

# Reduce context in config
export ZORD_CONTEXT_LENGTH=1024

Slow Performance

See OPTIMIZATION.md for tuning tips.

Contributing

Contributions are welcome! Please read our contributing guidelines before submitting PRs.

License

MIT License - See LICENSE for details.

Acknowledgments

• llama.cpp - GGUF quantization
• nanoGPT - Training philosophy
• MergeKit - Model merging
• DeepSeek-Coder - Base model
• Qwen2.5-Coder - Secondary model

---

Built with ❤️ by sajad