Skip to content

thinkingface

thinkingface is a self-hosted clone of the Hugging Face Hub. It hosts dataset and model repositories, versions them with git + Git LFS, keeps the actual bytes in your own Google Cloud Storage bucket, and tracks training runs — all behind an API that huggingface_hub and datasets already know how to talk to.

The thinkingface home page, showing instance totals and the most recently updated datasets and models

Who it is for

You already work with the Hugging Face Hub, and you need the same workflow somewhere else: on your own infrastructure, inside your own network, on storage you control. Point HF_ENDPOINT at a thinkingface instance and your existing Python code keeps working — no code changes, no vendored fork, no separate SDK.

What you get

  • Dataset and model repositories — create them from the web UI, the API, or the tf CLI, then git clone and git push over HTTP or SSH, with Git LFS for the large files.
  • A Hugging Face-compatible API — whoami, create_repo, upload_file, hf_hub_download, list_repo_files, load_dataset and friends work against your instance unchanged.
  • Browsing without downloading — a file tree, a Parquet table viewer, and a checkpoint metadata panel that reads only the header of a safetensors or PyTorch file.
  • Experiment tracking — a trackio-compatible interface records projects, runs, configs and metrics, and charts them in the web UI.
  • Organizations — shared team namespaces with admin / write / read roles.
  • Storage you can read directly — objects live in your bucket under content-addressed keys, and the web UI generates a gcloud storage cp script that restores any revision to its original file structure, locally or into another bucket.

Where to start

If you want to Go to
Get an instance running and upload a dataset Quickstart
Understand repositories, revisions and storage first Core Concepts
Push files from Python, the CLI, or git Uploading Files
Pull files back out, or read them straight from the bucket Downloading Files
Explore datasets and checkpoints in the browser Browsing the Web UI
Log training runs and compare them Tracking Experiments
Set up a shared team namespace Organizations
Trigger something of your own when a push or a run lands Webhooks
Know exactly what is and is not compatible with the Hub Compatibility
Deploy this for real Deployment

If you are evaluating rather than installing, Core Concepts is the shortest way to understand what thinkingface actually does with your data.