metadata

viewer: false
tags:
  - uv-script

Atlas Export

Generate and deploy interactive embedding visualizations to HuggingFace Spaces with a single command using Apple's Embedding Atlas library!

Quick Start

# Create a Space from any text dataset
uv run atlas-export.py stanfordnlp/imdb --space-name my-imdb-viz

# Your Space will be live at:
# https://huggingface.co/spaces/YOUR_USERNAME/my-imdb-viz

⚠️ Note: Currently this approach is limited by storage capacity for files on Spaces so for big datasets consider using the --sample option to limit the number of points visualized!

Examples

Image Datasets

# Visualize image datasets with CLIP
uv run atlas-export.py \
    beans \
    --space-name bean-disease-atlas \
    --image-column image \
    --model openai/clip-vit-base-patch32

Custom Embeddings

# Use a specific embedding model
uv run atlas-export.py \
    wikipedia \
    --space-name wiki-viz \
    --model nomic-ai/nomic-embed-text-v1.5 \
    --text-column text \
    --sample 50000

Pre-computed Embeddings

# If you already have embeddings in your dataset
uv run atlas-export.py \
    my-dataset-with-embeddings \
    --space-name my-viz \
    --no-compute-embeddings \
    --x-column umap_x \
    --y-column umap_y

GPU Acceleration (HF Jobs)

# First, get your HF token (if not already set)
python -c "from huggingface_hub import get_token; print(get_token())"

# Run on HF Jobs with GPU using experimental UV support
hf jobs uv run --flavor t4-small \
    -s HF_TOKEN=your-token-here \
    https://huggingface.co/datasets/uv-scripts/build-atlas/raw/main/atlas-export.py \
    stanfordnlp/imdb \
    --space-name imdb-viz \
    --model sentence-transformers/all-mpnet-base-v2 \
    --sample 10000

# With larger GPU and custom batch size for faster processing
hf jobs uv run --flavor a10g-large \
    -s HF_TOKEN=your-token-here \
    https://huggingface.co/datasets/uv-scripts/build-atlas/raw/main/atlas-export.py \
    your-dataset \
    --space-name your-atlas \
    --batch-size 64 \
    --sample 50000 \
    --text-column output

Note: Replace your-token-here with your actual token. Available GPU flavors: t4-small, t4-medium, l4x1, a10g-small, a10g-large.

Key Options

Option	Description	Default
`dataset_id`	HuggingFace dataset to visualize	Required
`--space-name`	Name for your Space	Required
`--model`	Embedding model to use	Auto-selected
`--text-column`	Column containing text	"text"
`--image-column`	Column containing images	None
`--sample`	Number of samples to visualize	All
`--batch-size`	Batch size for embedding generation	32 (text), 16 (images)
`--split`	Dataset split to use	"train"
`--local-only`	Generate locally without deploying	False
`--output-dir`	Local output directory	Temp dir
`--hf-token`	HuggingFace API token	From env/CLI

Run without arguments to see all options and more examples.

How It Works

Loads dataset from HuggingFace Hub
Generates embeddings (or uses pre-computed)
Creates static web app with embedded data
Deploys to HF Space

The resulting visualization runs entirely in the browser using WebGPU acceleration.

Credits

Built on Embedding Atlas by Apple. See the documentation for more details about the underlying technology.

Part of the UV Scripts collection 🚀