The Operating System for AI Research Labs

Designed for ML Researchers. Local, on-prem, or in the cloud. Open source.

⬇️ Install for Individuals · 🏢 Install for Teams · 📖 Documentation · 🎬 Demo · 💬 Discord

Transformer Lab Demo

--- ## ✨ What is Transformer Lab? Transformer Lab is an open-source machine learning platform that unifies the fragmented AI tooling landscape into a single, elegant interface. It is available in two editions:

### 👤 For Individuals **Perfect for researchers and hobbyists working on a single machine.** - **Local Privacy:** No data leaves your machine. - **Full Toolkit:** Train, fine-tune, chat, and evaluate models. - **Cross-Platform:** Runs natively on macOS (Apple Silicon), Linux, and Windows (WSL2). - **No Cloud Costs:** Use your own hardware.

### 🏢 For Teams **Built for research labs scaling across GPU clusters.** - **Unified Orchestration:** Submit jobs to **Slurm** clusters or **SkyPilot** clouds (AWS, GCP, Azure) from one UI. - **Collaborative:** Centralized experiment tracking, model registry, and artifact management. - **Interactive Compute:** One-click Jupyter, VSCode, and SSH sessions on remote nodes. - **Resilience:** Auto-recovery from checkpoints and spot instance preemption.

--- ## 🛠️ Key Capabilities

🧠 Foundation Models & LLMs

- **Universal Support:** Download and run Llama 3, DeepSeek, Mistral, Qwen, Phi, and more. - **Inference Engines:** Support for MLX, vLLM, Ollama, and HuggingFace Transformers. - **Format Conversion:** Seamlessly convert between HuggingFace, GGUF, and MLX formats. - **Chat Interface:** Multi-turn chat, batched querying, and function calling support.

🎓 Training & Fine-tuning

- **Unified Interface:** Train on local hardware or submit tasks to remote clusters using the same UI. - **Methods:** Full fine-tuning, LoRA/QLoRA, RLHF (DPO, ORPO, SIMPO), and Reward Modeling. - **Hardware Agnostic:** Optimized trainers for Apple Silicon (MLX), NVIDIA (CUDA), and AMD (ROCm). - **Hyperparameter Sweeps:** Define parameter ranges in YAML and automatically schedule grid searches.

🎨 Diffusion & Image Generation

- **Generation:** Text-to-Image, Image-to-Image, and Inpainting using Stable Diffusion and Flux. - **Advanced Control:** Full support for ControlNets and IP-Adapters. - **Training:** Train custom LoRA adaptors on your own image datasets. - **Dataset Management:** Auto-caption images using WD14 taggers.

📊 Evaluation & Analytics

- **LLM-as-a-Judge:** Use local or remote models to score outputs on bias, toxicity, and faithfulness. - **Benchmarks:** Built-in support for EleutherAI LM Evaluation Harness (MMLU, HellaSwag, GSM8K, etc.). - **Red Teaming:** Automated vulnerability testing for PII leakage, prompt injection, and safety.

🔌 Plugins & Extensibility

- **Plugin System:** Extend functionality with a robust Python plugin architecture. - **Lab SDK:** Integrate your existing Python training scripts (`import lab`) to get automatic logging, progress bars, and artifact tracking. - **CLI:** Power-user command line tool for submitting tasks and monitoring jobs without a browser.

🗣️ Audio Generation

- **Text-to-Speech:** Generate speech using Kokoro, Bark, and other state-of-the-art models. - **Training:** Fine-tune TTS models on custom voice datasets.

--- ## 📥 Quick Start ### 1. Install ```bash curl https://lab.cloud/install.sh | bash ``` ### 2. Run ```bash cd ~/.transformerlab/src ./run.sh ``` ### 3. Access Open your browser to `http://localhost:8338`. #### Requirements | Platform | Requirements | |----------|-------------| | **macOS** | Apple Silicon (M1/M2/M3/M4) | | **Linux** | NVIDIA or AMD GPU | | **Windows** | NVIDIA GPU via WSL2 ([setup guide](https://lab.cloud/docs/install/windows-wsl-cuda)) | --- ## 🏢 Enterprise & Cluster Setup Transformer Lab for Teams runs as an overlay on your existing infrastructure. It does not replace your scheduler; it acts as a modern control plane for it. To configure Transformer Lab to talk to **Slurm** or **SkyPilot**: 1. Follow the [Teams Install Guide](https://lab.cloud/for-teams/install). 2. Configure your compute providers in the Team Settings. 3. Use the CLI (`lab`) or Web UI to queue tasks across your cluster. --- ## 👩‍💻 Development

Frontend

```bash # Requires Node.js v22 npm install npm start ```

Backend (API)

```bash cd api ./install.sh # Sets up Conda env + Python deps ./run.sh # Start the API server ```

Lab SDK

```bash pip install transformerlab ```

--- ## 🤝 Contributing We are an open-source initiative backed by builders who care about the future of AI research. We welcome contributions! Please check our [issues](https://github.com/transformerlab/transformerlab-app/issues) for open tasks.

--- ## 📄 License AGPL-3.0 · See [LICENSE](LICENSE) for details. --- ## 📚 Citation ```bibtex @software{transformerlab, author = {Asaria, Ali and Salomone, Tony}, title = {Transformer Lab: The Operating System for AI Research}, year = 2023, url = {https://github.com/transformerlab/transformerlab-app} } ``` --- ## 💬 Community

_{Built with ❤️ by Transformer Lab in Canada 🇨🇦}