🍜 Soup - Train AI Models on Your Own Computer

🚀 What is Soup?

Soup is a simple tool that lets you teach your own AI model using just one settings file. You don't need to be a programmer or have expensive equipment. If you have a normal laptop, you can create a custom AI assistant that understands your specific needs.

Think of it like this: you're baking a cake. The recipe is your settings file. Soup takes that recipe and uses your computer's graphics card (GPU) to bake the cake - which is a working AI model you can talk to.

🎯 Who Should Use Soup?

• Students who want to learn how AI works
• Hobbyists who want a personal AI with their own data
• Small business owners who need a custom chatbot
• Curious people who want to experiment with AI

No coding experience is required. If you can edit a text file and click a button, you can use Soup.

✨ Key Features

• One File Setup - Write all your settings in a single YAML file
• Laptop Friendly - Works on computers with only 4 GB of video memory
• Powerful Results - Can create models with 8 billion parameters (think of parameters as "brain cells" - more means smarter)
• Streaming Technology - Processes data in chunks, so your computer doesn't choke
• Works with Popular Tools - Compatible with Ollama, Hugging Face, and other AI tools
• Multiple Training Methods - Supports advanced techniques like QLoRA and DPO

📦 Download and Install

Visit this link to download the application.

Once you're on the download page:

1. Look for the latest version at the top
2. Choose the file that matches your computer (Windows, Mac, or Linux)
3. Click the download button

🖥️ System Requirements

• Operating System: Windows 10/11, macOS 12+, or modern Linux
• Memory: 8 GB RAM (16 GB recommended)
• Graphics Card: NVIDIA GPU with at least 4 GB VRAM
• Storage: 20 GB free space (AI models take up room)
• Internet: Needed for downloading and updating

🛠️ Getting Started

Step 1: Create Your Settings File

Make a new text file called config.yaml. Here's a simple example:

model:
  name: "my-custom-ai"
  base_model: "mistral-7b"
  training:
    method: "qlora"
    epochs: 3
    batch_size: 2
    learning_rate: 0.0002

Step 2: Add Your Data

Create a folder called data and put your training examples there. Each file should have question-answer pairs like:

{"question": "What is the return policy?", "answer": "Returns are accepted within 30 days."}

Step 3: Run Soup

Open your terminal or command prompt and navigate to the Soup folder:

python soup.py config.yaml

That's it! Soup will start training your AI. You'll see a progress bar showing how far along it is.

📚 How Soup Works

Think of Soup as a teacher for AI models. Here's what happens:

1. Preparation - Soup reads your settings and gathers your data
2. Training - Your computer's GPU processes examples in small batches
3. Refinement - The model learns patterns from your data
4. Export - You get a ready-to-use AI file

The clever part is the "layered streaming" - this means your computer doesn't try to handle everything at once. It processes tiny pieces at a time, which is why it works on low-memory machines.

🔧 Training Methods Explained

• SFT (Supervised Fine-Tuning) - Best for teaching specific skills
• DPO (Direct Preference Optimization) - Great for making your AI helpful and honest
• QLoRA - The most memory-efficient method, perfect for small GPUs
• LoRA - A lightweight approach that monitors only the most important parts

💡 Real-World Uses

You can create AI models for:

• Customer service - Answer common questions about your products
• Education - Build a tutor that teaches your subject
• Writing help - Train an AI in your writing style
• Data organization - Make sense of your company documents
• Fun - Create a chatbot with a personality you design

🌐 Compatibility

Soup works seamlessly with:

• Ollama - Run your trained models locally
• Hugging Face - Share and download models
• GGUF format - Lightweight model format for smaller devices
• Transformers - Use with popular AI libraries
• PyTorch - The underlying framework powering everything

❓ Frequently Asked Questions

Q: Do I need to know Python? A: No. You only need to edit a text file and run a command.

Q: How long does training take? A: With a 4 GB GPU, expect 2-4 hours for a simple model. Your settings affect this.

Q: Can I use an AMD or Intel GPU? A: The current version works best with NVIDIA. Other GPUs may work but aren't officially supported.

Q: What happens if my computer runs out of memory? A: Soups streaming technology automatically adjusts. You might see slower training, but it shouldn't crash.

Q: Is my data safe? A: Everything stays on your computer. Nothing is uploaded anywhere.

🚨 Troubleshooting

Training stops with "CUDA out of memory"

• Close other programs using your GPU
• Reduce batch_size in your config
• Try using a smaller base model

Can't find the YAML file

• Make sure the file extension is .yaml not .txt
• Check the file is in the same folder as your soup.py

Model talks nonsense after training

• Add more examples to your data
• Increase epochs in your config
• Check your data for errors

📖 Getting Help

• Read the detailed documentation in the docs folder
• Check the GitHub Issues for known problems
• Join our community discussions

🤝 Contributing

Want to improve Soup? We welcome:

• Bug reports
• Feature suggestions
• Documentation improvements
• Example configurations

Visit the GitHub repository to learn more.

📄 License

Soup is released under the MIT License - you can use, modify, and share it freely.

---

Keywords: cli, consumer-gpu, dpo, fine-tuning, gguf, huggingface, llm, llmops, local-ai, local-llm, lora, low-vram, machine-learning, ollama, peft, python, pytorch, qlora, sft, transformers