Marimo UV Scripts

Marimo notebooks that work as both interactive tutorials and batch scripts.

What is this?

Marimo notebooks are pure Python files that can be:

• Edited interactively with a reactive notebook interface
• Run as scripts with uv run - same as any UV script

This makes them perfect for tutorials and educational content where you want users to explore step-by-step, but also run the whole thing as a batch job.

Available Scripts

Script	Description
getting-started.py	Introduction to UV scripts and HF datasets
train-image-classifier.py	Fine-tune a Vision Transformer on image classification
_template.py	Minimal template for creating your own notebooks

Usage

Run as a script

# Get dataset info
uv run https://huggingface.co/datasets/uv-scripts/marimo/raw/main/getting-started.py --dataset squad

# Train an image classifier
uv run https://huggingface.co/datasets/uv-scripts/marimo/raw/main/train-image-classifier.py \
    --dataset beans \
    --epochs 3 \
    --output-repo your-username/beans-vit

Run interactively

# Clone and edit locally
git clone https://huggingface.co/datasets/uv-scripts/marimo
cd marimo

# Open in marimo editor (--sandbox auto-installs dependencies)
uvx marimo edit --sandbox getting-started.py
uvx marimo edit --sandbox train-image-classifier.py

Run on HF Jobs (GPU)

# Train image classifier with GPU
hf jobs uv run --flavor l4x1 --secrets HF_TOKEN \
    https://huggingface.co/datasets/uv-scripts/marimo/raw/main/train-image-classifier.py \
    -- --dataset beans --output-repo your-username/beans-vit --epochs 5 --push-to-hub

Why Marimo?

• Reactive: Cells automatically re-run when dependencies change
• Pure Python: No JSON, git-friendly, readable as plain code
• Self-contained: Inline dependencies via PEP 723 metadata
• Dual-mode: Same file works as notebook and script

Create Your Own Marimo UV Script

Use _template.py as a starting point:

# Clone and copy the template
git clone https://huggingface.co/datasets/uv-scripts/marimo
cp marimo/_template.py my-notebook.py

# Edit interactively
uvx marimo edit --sandbox my-notebook.py

# Test as script
uv run my-notebook.py --help

Recipes

Add explanation (notebook only)

mo.md("""
## This is a heading

This text explains what's happening. Only shows in interactive mode.
""")

Show output in both modes

# print() shows in terminal (script) AND cell output (notebook)
print(f"Loaded {len(data)} items")

Interactive control with CLI fallback

# Parse CLI args first
parser = argparse.ArgumentParser()
parser.add_argument("--count", type=int, default=10)
args, _ = parser.parse_known_args()

# Create UI control with CLI default
slider = mo.ui.slider(1, 100, value=args.count, label="Count")

# Use it - works in both modes
count = slider.value  # UI value in notebook, CLI value in script

Show visuals (notebook only)

# mo.md() with images, mo.ui.table(), etc. only display in notebook
mo.ui.table(dataframe)

# For script mode, also print summary
print(f"DataFrame has {len(df)} rows")

Conditional notebook-only code

# Check if running interactively
if hasattr(mo, 'running_in_notebook') and mo.running_in_notebook():
    # Heavy visualization only in notebook
    show_complex_plot(data)

Best Practices

1. Always include print() for important output - It works in both modes
2. Use argparse for all configuration - CLI args work everywhere
3. Add mo.md() explanations between steps - Makes tutorials readable
4. Test in script mode first - Ensure it works without interactivity
5. Keep dependencies minimal - Add marimo plus only what you need

Learn More

• Marimo documentation
• UV scripts guide
• uv-scripts organization