HuggingFace Integration

Hubify Labs integrates with HuggingFace Hub for publishing models, datasets, and experiment artifacts. Share your research outputs with the community directly from your lab.

Setup

Get a HuggingFace token

Go to huggingface.co/settings/tokens and create a write-access token.

Configure in Hubify

hubify config set huggingface-token "hf_abc123..."

Verify

hubify config test huggingface

HuggingFace connection: OK
User: houston-golden
Organizations: hubify-labs

Publishing Datasets

Push experiment outputs and processed datasets to HuggingFace:

# Push a dataset from an experiment
hubify data publish EXP-054 \
  --target huggingface \
  --repo hubify-labs/planck-bao-chains \
  --description "MCMC posterior samples from Planck+BAO analysis"

This creates a HuggingFace dataset repository with:

• Data files (Parquet format for tabular data)
• Dataset card (auto-generated from experiment metadata)
• Schema and column descriptions
• Provenance information (experiment ID, lab, date)

Dataset Cards

Hubify auto-generates HuggingFace dataset cards:

hubify data card EXP-054

The card includes:

• Dataset description and purpose
• Column definitions
• Collection methodology
• License information
• Citation instructions

Publishing Models

Push trained models to HuggingFace:

hubify model publish \
  --experiment EXP-060 \
  --repo hubify-labs/anomaly-detector-v2 \
  --framework pytorch

Publishing Spaces

Create a HuggingFace Space with an interactive demo:

hubify space create \
  --repo hubify-labs/bounce-cosmology-explorer \
  --type gradio \
  --experiment EXP-054

This creates a Gradio app that lets anyone explore your experiment results interactively.

Bulk Publishing

Publish all publishable artifacts from a lab:

# List what can be published
hubify data list --publishable

# Publish everything
hubify data publish --all --target huggingface --org hubify-labs

Linking to Lab Site

Published HuggingFace resources are automatically linked from your lab site:

# After publishing, sync the site
hubify site sync

The lab site will display links to your HuggingFace datasets, models, and Spaces.

Examples

# Publish anomaly catalog
hubify data publish EXP-045 \
  --target huggingface \
  --repo hubify-labs/desi-anomaly-catalog \
  --description "195,829 anomalous spectra from DESI DR1"

# Publish MCMC chains for reproducibility
hubify data publish EXP-054 \
  --target huggingface \
  --repo hubify-labs/bounce-cosmology-mcmc \
  --description "424K posterior samples across 4 dataset combinations"

# Publish a trained autoencoder
hubify model publish \
  --experiment EXP-038 \
  --repo hubify-labs/spectral-anomaly-ae \
  --framework pytorch \
  --description "Autoencoder for spectral anomaly detection (16D latent)"

Privacy

• Only experiments and data you explicitly publish are uploaded
• Private lab data is never automatically shared
• You control the repository visibility (public or private) on HuggingFace
• API tokens are stored locally and never sent to Hubify servers