TinyTorch/binder

Binder Environment Setup

This directory contains configuration files for running TinyTorch in cloud environments via Binder and Google Colab.

Files

• requirements.txt: Python dependencies for the Binder environment
• postBuild: Script that runs after environment setup to install TinyTorch

How It Works

Binder

When users click the "Launch Binder" button on any notebook page in the TinyTorch documentation:

1. Binder reads binder/requirements.txt to install Python dependencies
2. Binder runs binder/postBuild which:
   • Installs the TinyTorch package (pip install -e .)
   • Generates student notebooks from src/*.py files using Jupytext
   • Populates modules/ with ready-to-use Jupyter notebooks
3. Users get a fully configured JupyterLab environment with TinyTorch and all notebooks ready to use

Note: The modules/ directory is gitignored because notebooks are generated from the source .py files. This ensures students always get notebooks that match the current code.

Binder URL Format:

https://mybinder.org/v2/gh/mlsysbook/TinyTorch/main

Google Colab

Colab launch buttons automatically:

1. Clone the repository
2. Install dependencies from binder/requirements.txt
3. Run setup commands (users may need to manually run pip install -e .)

Colab URL Format:

https://colab.research.google.com/github/mlsysbook/TinyTorch/blob/main/path/to/notebook.ipynb

Testing

To test your Binder setup:

1. Test Binder Build:

   # Visit: https://mybinder.org/v2/gh/mlsysbook/TinyTorch/main
   # Or use the badge:
   [![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/mlsysbook/TinyTorch/main)
   

2. Verify Installation: Once Binder launches, test in a notebook:

   import tinytorch
   print(tinytorch.__version__)
   

3. Check Available Resources:

   import os
   print("Modules:", os.listdir("modules"))
   print("Assignments:", os.listdir("assignments"))
   print("Milestones:", os.listdir("milestones"))
   

Troubleshooting

Binder Build Fails

• Check binder/requirements.txt for syntax errors
• Verify binder/postBuild has execute permissions (chmod +x binder/postBuild)
• Review Binder build logs at: https://mybinder.org/v2/gh/mlsysbook/TinyTorch/main?urlpath=lab/tree/logs%2Fbuild.log

Colab Import Errors

• Ensure binder/requirements.txt includes all dependencies
• Users may need to run: !pip install -e . in a Colab cell
• Check that the repository is public (Colab can't access private repos)

Package Not Found

• Verify postBuild script runs pip install -e . correctly
• Check that pyproject.toml is in the repository root
• Ensure all dependencies in requirements.txt are compatible

Deployment Environments

As documented in the TinyTorch paper, three deployment environments are supported:

1. JupyterHub (institutional server)

   • 8-core/32GB supports ~50 students
   • Best for classroom use

2. Google Colab (zero installation)

   • Best for MOOCs and self-paced learning
   • No setup required from students

3. Local Installation (pip install tinytorch)

   • Best for self-paced learning and development
   • Full control over environment

Keeping Dependencies Updated

When updating dependencies:

1. Update requirements.txt (root) - for local development
2. Update binder/requirements.txt - for Binder/Colab
3. Update docs/requirements.txt - for documentation builds
4. Keep versions synchronized where possible

References

• Binder Documentation
• Jupyter Book Launch Buttons
• Google Colab GitHub Integration