transformerlab-app: an open-source machine learning platform that unifies AI research tooling and cluster orchestration

What it solves

Transformer Lab is designed to unify the fragmented landscape of AI tooling. It provides a single interface for ML researchers to manage the entire lifecycle of AI models—from training and fine-tuning to inference and evaluation—across local machines, on-prem clusters, and cloud environments.

How it works

The platform acts as a control plane that integrates with various inference engines (like vLLM, Ollama, and MLX) and compute schedulers (such as Slurm and SkyPilot). It supports multiple hardware backends including Apple Silicon, NVIDIA, and AMD GPUs. For individuals, it runs locally for privacy; for teams, it provides centralized orchestration, experiment tracking, and interactive compute sessions (Jupyter, VSCode) on remote nodes.

Who it’s for

It is built for ML researchers, hobbyists, and AI research labs that need a streamlined way to coordinate training tasks, manage model registries, and run evaluations across diverse hardware.

Highlights

• Unified AI Toolkit: Support for LLMs, Diffusion models, and Text-to-Speech models in one UI.
• Collaborative Orchestration: Submit jobs to Slurm or SkyPilot clusters with auto-recovery from checkpoints.
• Comprehensive Training: Supports full fine-tuning, LoRA/QLoRA, RLHF (DPO, ORPO, SIMPO), and hyperparameter sweeps.
• Built-in Evaluation: Includes LLM-as-a-Judge, red teaming for safety, and integration with the EleutherAI LM Evaluation Harness.
• Extensible Architecture: A Python plugin system and Lab SDK allow users to integrate existing training scripts with automatic logging.