The Universal Command-Line Tool for LLM Fine-Tuning
Scale from RTX 4060 to H200 clusters. Train any model, any size, anywhere.

Xyne-Play is a production-ready CLI tool that makes LLM fine-tuning accessible to everyone. Whether you're a hobbyist with a consumer GPU or an enterprise with GPU clusters, Xyne-Play provides one unified interface for:

• 🏠 Hobbyists: Fine-tune 7B models on RTX 4060 (8GB)
• 🎓 Researchers: Experiment with 13B-20B models on RTX 4090 (24GB)
• 🏢 Enterprises: Train 70B+ models on H100/H200 clusters
• 🚀 Everyone: One tool, any hardware, any model size

git clone https://github.com/xynehq/play
cd play
make install && make setup-dirs

• Python: 3.8+ (3.10+ recommended)
• CUDA: 11.8+ for GPU training
• Memory: 16GB+ RAM recommended
• Storage: 50GB+ for models and data

# Interactive setup with sample data
./workflows/quick_start.sh

# Start training with monitoring
make train-bnb-tb

# Chat with your model
make infer

Create data/raw/my_data.jsonl:

{"system": "You are a helpful assistant.", "user": "What is machine learning?", "assistant": "Machine learning is..."}
{"system": "You are a helpful assistant.", "user": "Explain neural networks.", "assistant": "Neural networks are..."}

# Process data and start training
make process && make train-bnb-tb

# Interactive chat with trained model
make infer

# Use custom config
make train CONFIG=configs/my_config.yaml

# Override specific parameters
make train CONFIG=configs/run_bnb.yaml model.name=Qwen/Qwen2.5-7B-Instruct

• 📋 Documentation - Complete guides and tutorials
• 🎯 Task-Specific Examples - SFT, DAPT, Embedding FT, and more
• 🚀 Advanced Guides - Multi-GPU, DeepSpeed, production deployment
• 🔧 Configuration Reference - All configuration options

make help           # See all available commands
make gpu-info       # Check your hardware
make check          # Validate setup
make train-bnb-tb   # Start training with monitoring
make infer          # Chat with your model

Apache License 2.0 - see LICENSE for details.

Xyne-Play: From 8GB to 800GB. From hobbyist to enterprise. One tool, infinite possibilities. 🚀