tunestudio 0.0.11

An all in one platform for finetuning your LLM models using your Google Colab, Kaggle or Local resources.

TuneStudio - No-Code LLM Fine-Tuning Platform

A beautiful, modern web interface for fine-tuning large language models without writing code. Built with Flask backend and React frontend, designed to run locally or in cloud environments like Google Colab and Kaggle.

✨ Features

• 🚀 No-Code Interface: Beautiful, modern web UI for configuring and monitoring fine-tuning jobs
• 🤖 50+ Pre-supported Models: Including Llama 3.1, Gemma, Phi-3, Mistral, Qwen2, BERT, and more
• ⚡ Real-time Logging: Live progress monitoring with automatic log updates
• 🔧 Flexible Configuration: Easy hyperparameter tuning with sensible defaults
• ☁️ Cloud-Ready: Works seamlessly in Colab/Kaggle with ngrok integration
• 💻 Local Execution: Uses your local GPU/CPU resources - no data sent to external services
• 📊 Professional UI: Dark theme, responsive design, smooth animations
• 🔄 Auto-scrolling Logs: Real-time log updates with automatic scroll-to-bottom

🚀 Quick Start

Installation

pip install tunestudio

text

Local Usage

Start the web interface tunestudio

text

Then open your browser to http://localhost:5001

Google Colab / Kaggle Usage

In a Colab/Kaggle notebook cell import tunestudio tunestudio.main.run_app_with_ngrok()

text

Or use the command line: tunestudio-colab

text

📖 Usage Guide

1. Dataset Preparation

• Local CSV File: Place your training data in a CSV file with a text column
• Hugging Face Dataset: Use any dataset name from Hugging Face Hub (e.g., imdb, rotten_tomatoes)

Example CSV format: text "This is a positive example for training." "This is another training sample." "Fine-tuning is made easy with TuneStudio."

text

2. Model Selection

Choose from 50+ supported models including:

• Llama 3.1: meta-llama/Llama-3.1-8B, meta-llama/Llama-3.1-8B-Instruct
• Gemma: google/gemma-2b, google/gemma-7b, google/gemma-2-9b
• Phi-3: microsoft/Phi-3-mini-4k-instruct, microsoft/Phi-3-small-8k-instruct
• Mistral: mistralai/Mistral-7B-v0.1, mistralai/Mixtral-8x7B-v0.1
• And many more...

3. Configuration

1. Enter your dataset path or Hub name
2. Select a model from the dropdown
3. Adjust hyperparameters (optional):
   • Learning Rate (default: 2e-5)
   • Batch Size (default: 16)
   • Epochs (default: 3)

4. Training

1. Click "Start Fine-Tuning"
2. Monitor real-time progress in the live log panel
3. Your fine-tuned model will be saved to ./fine_tuned_model/

🛠️ Supported Models

Large Language Models

• Llama 3.1: All variants (8B, 70B, 405B, Instruct versions)
• Gemma: Google's efficient models (2B, 7B, 9B, 27B variants)
• Phi-3: Microsoft's small but powerful models (mini, small, medium)
• Mistral: Including Mixtral MoE variants (7B, 8x7B, 8x22B)
• Qwen2: Alibaba's multilingual models (0.5B to 72B)

Encoder Models (for classification)

• BERT Family: bert-base-uncased, bert-large-uncased, distilbert-base-uncased
• RoBERTa: roberta-base, roberta-large
• ALBERT: albert-base-v2, albert-large-v2

Other Popular Models

• GPT: EleutherAI/gpt-neo-125M, EleutherAI/gpt-j-6B
• BLOOM: bigscience/bloom-560m, bigscience/bloom-1b7
• Falcon: tiiuae/falcon-7b, tiiuae/falcon-7b-instruct

🔧 Requirements

System Requirements

• Python: 3.8 or higher
• RAM: 4GB minimum (16GB+ recommended for larger models)
• GPU: Recommended but not required (CUDA-compatible for faster training)
• Storage: At least 10GB free space for model downloads

Dependencies

All dependencies are automatically installed:

• PyTorch
• Transformers
• Flask & Flask-CORS
• Datasets
• Accelerate
• Pandas
• Hugging Face Hub
• pyngrok (for Colab/Kaggle)

🌐 Cloud Platforms

Google Colab

Install and run in a Colab cell !pip install tunestudio import tunestudio tunestudio.main.run_app_with_ngrok()

text

Kaggle Notebooks

Install and run in a Kaggle cell import subprocess subprocess.run(["pip", "install", "tunestudio"]) import tunestudio tunestudio.main.run_app_with_ngrok()

text

🚨 Authentication for Gated Models

Some models (like Llama) require Hugging Face authentication:

Login to Hugging Face huggingface-cli login

text

Or set your token as an environment variable: export HUGGINGFACE_HUB_TOKEN=your_token_here

text

🔍 Troubleshooting

Common Issues

1. Model Not Loading

• Ensure you have sufficient RAM/VRAM
• For gated models, make sure you're authenticated with Hugging Face
• Try a smaller model first (e.g., distilbert-base-uncased)

2. Dataset Loading Failed

• Check file path is correct for local files
• Ensure CSV has a text column
• For Hub datasets, check the dataset name is valid

3. CUDA Out of Memory

• Reduce batch size (try 8 or 4)
• Use a smaller model
• Enable gradient checkpointing (automatically handled)

4. Slow Training

• Consider using a GPU if available
• Reduce the number of epochs
• Use a smaller model for testing

📊 Performance Tips

• GPU Usage: Training is significantly faster with a CUDA-compatible GPU
• Batch Size: Start with 16, reduce if you encounter memory issues
• Model Size: Begin with smaller models (2B-7B parameters) for faster iteration
• Dataset Size: Larger datasets require more time but generally produce better results

🤝 Contributing

Contributions are welcome! Here's how you can help:

1. Fork the repository
2. Create a feature branch: git checkout -b feature-name
3. Make your changes and add tests
4. Submit a pull request

Development Setup

git clone https://github.com/yourusername/tunestudio.git cd tunestudio pip install -e .

text

📄 License

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

🙏 Acknowledgments

• Built with Transformers by Hugging Face
• UI powered by React and Tailwind CSS
• Backend built with Flask

📞 Support

• Issues: GitHub Issues
• Documentation: Wiki
• Email: [e-mail]

---

Made with ❤️ for the AI community