transformerlab-app

Why we included this project

Transformer Lab bundles fine-tuning, evaluation, and inference into one application that runs on your own machine or sends jobs to a Slurm or SkyPilot cluster. You load models like Llama, DeepSeek, Mistral, or Stable Diffusion, train LoRA or QLoRA adapters on Apple Silicon or NVIDIA GPUs, and run built-in benchmarks such as MMLU or LLM judge scoring from the same interface. The individual edition keeps everything local, so no training data leaves your hardware, which matters for labs with privacy constraints and for hobbyists who just want to experiment on a single GPU. A Lab SDK and plugin system lets you reuse existing Python training scripts and get automatic logging and artifact tracking without rewriting them.