transformerlab

transformerlab-app: Train and Evaluate AI Models in One Workspace

Transformer Lab is a research workspace for training, fine-tuning, running, and evaluating AI models on local machines or GPU clusters. It suits individual researchers who want a unified UI and teams that need to coordinate jobs across existing infrastructure.

View on GitHub Compare
Stars
5.2k
Forks
551
Stars, last 30 days
+1

Our take

Health score

55/100

How it is scored
Activity
75
Community
64
Issues
39
Pull requests
43

+1 stars in the last 30 days

  • Releases more often than 85% of the projects we track
  • Slower to close issues than 90% of the projects we track

Project history

Daily GitHub snapshots since Oct 7, 2026. History covers stars and forks; health signals are not tracked historically.

Selected snapshot

Stars

5,198

Forks

552

Change since Nov 14, 2026: 0 stars·-1 forks

Overview

Transformer Lab brings model work that often spans separate tools into one research environment. It provides a UI and related tools for running models, fine-tuning them, generating images or audio, and evaluating outputs.

It supports individual setups as well as team workflows that submit work to Slurm clusters or SkyPilot cloud infrastructure. This can be useful when researchers want one interface across local and remote compute, but it does not replace the underlying scheduler or hardware.

Key Features

  • Run and chat with models using MLX, vLLM, Ollama, or Hugging Face Transformers.
  • Fine-tune models with full training, LoRA/QLoRA, and preference optimization methods including DPO, ORPO, and SIMPO.
  • Submit jobs through a shared interface to local hardware, Slurm, or SkyPilot.
  • Generate images with Stable Diffusion and Flux, with ControlNet and IP-Adapter support.
  • Evaluate model outputs with benchmarks, model-based judging, and automated red-teaming tools.
  • Extend workflows with Python plugins, a Lab SDK, and a command-line tool.
  • Generate and fine-tune text-to-speech models.

Use Cases

  • An individual researcher can run and fine-tune models on a supported workstation without sending data to a cloud service.
  • An ML team can launch experiments on Slurm or SkyPilot while using a common interface for job submission and artifact tracking.
  • A model developer can compare checkpoints with built-in benchmarks or model-based evaluation before deployment.
  • A generative-media researcher can train image-generation LoRAs or experiment with image and speech generation.

Getting Started

Install with the README's quick-start command:

curl https://lab.cloud/install.sh | bash

Then run cd ~/.transformerlab/src && ./run.sh and open http://localhost:8338. See the README for platform requirements, team setup, and development instructions.

Alternatives

  • xTuring: xTuring is a Python library for model customization and inference, rather than a unified research workspace for managing training and evaluation jobs.
  • LLMBox: LLMBox focuses on configurable training and benchmark evaluation pipelines, rather than a broader workspace for running models and coordinating jobs.
ProjectLanguageLicenseStarsStatus
transformerlab-appPythonAGPL-3.05.2kNot checked yet
xTuringPythonApache-2.02.7kActive
LLMBoxPythonMIT848Inactive

Considerations

  • Compute requirements depend on the workload. The README lists Apple Silicon for macOS, NVIDIA or AMD GPUs for Linux, and NVIDIA via WSL2 for Windows.
  • Team deployment connects to Slurm or SkyPilot; it acts as a control plane and does not replace the existing scheduler.
  • The AGPL-3.0 license should be reviewed before incorporating or modifying the software in a service or product.
  • The project combines a Python API with a frontend, so self-hosting and development involve more than a standalone Python package.

Found this useful?

Share it with someone who would like transformerlab-app.

Comparisons