Transformer Lab
An open-source machine learning platform that unifies training, fine-tuning, chat and evaluation of models into one interface, running locally or scaled to GPU clusters.
Transformer Lab offers an individual edition for researchers and hobbyists on a single machine, with local privacy, no cloud cost, and cross-platform support on macOS, Linux and Windows via WSL2, and a team edition for research labs that submits jobs to Slurm clusters or SkyPilot clouds from one UI, with centralized experiment tracking, a model registry, one-click Jupyter, VS Code and SSH sessions, and auto-recovery from checkpoints or spot preemption.
It downloads and runs models such as Llama 3, DeepSeek, Mistral, Qwen and Phi through MLX, vLLM, Ollama or Hugging Face Transformers, converts between Hugging Face, GGUF and MLX formats, and supports full fine-tuning, LoRA/QLoRA, RLHF methods such as DPO, ORPO and SIMPO, and reward modeling with hardware-specific trainers for Apple Silicon, NVIDIA and AMD. It also generates images with Stable Diffusion and Flux, trains LoRA adapters on custom image datasets, auto-captions images, and evaluates models with an LLM acting as judge.
It is licensed under AGPL-3.0.