visu-hlo 1.0.0b1

Displays the HLO representation of (un-)jitted functions as SVG

Visualizing HLO computation graphs

visu-hlo is a Python package that displays the HLO (High Level Operations) representation of JAX functions as SVG visualizations. This tool helps developers understand the computational graphs generated by JAX compilation.

Features

• 🎯 Easy Visualization: Display HLO graphs with a single function call
• ⚡ JIT Support: Works with both regular and jitted JAX functions
• 🖼️ SVG Output: High-quality vector graphics that scale perfectly
• 🖥️ Cross-Platform: Supports Linux, macOS, and Windows
• 📦 Lightweight: Minimal dependencies - just JAX and Graphviz

Quick Example

import jax.numpy as jnp
from visu_hlo import show

def func(x):
    return 3 * x * 2

# Display optimized HLO (default)
show(func, jnp.ones(10))

To display the non-optimized HLO:

show(func, jnp.ones(10), jit=False)

To save as an SVG file:

from visu_hlo import write_svg
write_svg('graph.svg', func, jnp.ones(10))

Installation

pip install visu-hlo

System dependency: Install Graphviz

Documentation

Full documentation: https://visu-hlo.readthedocs.io/

License

MIT