Skip to content

PYODIDE_USAGE.md

Latest commit

 

History

History
193 lines (134 loc) · 4.57 KB

PYODIDE_USAGE.md

File metadata and controls

193 lines (134 loc) · 4.57 KB
Raw

Using PythonCK in Pyodide/Quarto

This document explains how to use the pythonck package in Pyodide environments, including Quarto documents.

Package Overview

The pythonck package provides Python implementations of Composable Kernels tensor operations, organized into three main modules:

  • pytensor: Core tensor operations, coordinates, descriptors, and views
  • tensor_transforms: Tensor transformation parsing and analysis
  • tile_distribution: Tile distribution algorithms and visualization

Installation in Pyodide

Option 1: Install from Wheel

If you have uploaded the wheel to a web server or PyPI:

import micropip
await micropip.install("pythonck")

Option 2: Install from Local Wheel

If you have the wheel file locally:

import micropip
await micropip.install("path/to/pythonck-0.1.0-py3-none-any.whl")

Option 3: Install from URL

If the wheel is hosted online:

import micropip
await micropip.install("https://example.com/path/to/pythonck-0.1.0-py3-none-any.whl")

Usage in Quarto

Basic Setup

# In a Quarto Python cell
import micropip
await micropip.install("pythonck")  # or your wheel URL

# Import the modules
import pytensor
import tensor_transforms
import tile_distribution

Example: Tensor Coordinates

from pytensor import make_tensor_coordinate, make_tensor_adaptor_coordinate

# Create a simple tensor coordinate
coord = make_tensor_coordinate([64, 128])
print(f"Tensor shape: {coord.get_lengths()}")

# Create an adaptor coordinate for transformations
adaptor_coord = make_tensor_adaptor_coordinate([32, 64], [0, 0])
print(f"Adaptor coordinate at origin: {adaptor_coord.get_offset()}")

Example: Tensor Transforms

from tensor_transforms import TensorTransformParser

# Parse tensor transform expressions
parser = TensorTransformParser()
result = parser.parse_expression("Transform<Merge<1, 2>>")
print(f"Parsed transform: {result}")

Example: Tile Distribution

from tile_distribution import get_default_variables, TileDistributionVisualizer

# Get default configuration
config = get_default_variables()
print(f"Default tile configuration: {config}")

# Create visualizations (if matplotlib is available)
try:
    visualizer = TileDistributionVisualizer()
    # Use visualizer for tile distribution plots
except ImportError:
    print("Matplotlib not available for visualization")

Dependencies

The package has minimal dependencies for Pyodide compatibility:

  • Required: sympy>=1.9 (for symbolic mathematics)
  • Optional: matplotlib, numpy, pandas (for visualization features)

Pyodide Compatibility Notes

  1. Pure Python: The package is 100% pure Python, ensuring full Pyodide compatibility
  2. No Native Extensions: No C extensions or compiled code
  3. Minimal Dependencies: Only requires sympy for core functionality
  4. Wheel Format: Distributed as a universal wheel (py3-none-any)

Building the Wheel

To build the wheel yourself:

# Install build tools
pip install build

# Build the wheel
python -m build --wheel

# The wheel will be created in dist/pythonck-0.1.0-py3-none-any.whl

Example Quarto Document

Here's a complete example for a Quarto document:

---
title: "PythonCK in Quarto"
format: html
jupyter: python3
---

```{python}
import micropip
await micropip.install("pythonck")  # Replace with your wheel URL
# Import and use pythonck
import pytensor
from tensor_transforms import TensorTransformParser

# Create a tensor coordinate
coord = pytensor.make_tensor_coordinate([128, 256])
print(f"Created tensor with shape: {coord.get_lengths()}")

# Parse a transform
parser = TensorTransformParser()
transform = parser.parse_expression("Transform<PassThrough<0, 1>>")
print(f"Parsed transform: {transform}")


## Troubleshooting

### Import Errors

If you encounter import errors, ensure:

1. The package was installed successfully
2. All dependencies are available in Pyodide
3. The correct module names are used

### SymPy Issues

If SymPy-related functions fail:

```python
# Check SymPy installation
import sympy
print(f"SymPy version: {sympy.__version__}")

Memory Limitations

For large tensor operations in Pyodide:

  • Use smaller tensor sizes for demonstrations
  • Consider pagination for large visualizations
  • Monitor browser memory usage

API Reference

For detailed API documentation, see:

  • pytensor/README.md - Core tensor operations
  • pytensor/IMPLEMENTATION_STATUS.md - Implementation status
  • Module docstrings and type hints throughout the codebase