Grady

This project is the basic implementation of a Deep Learning Framework, following the concepts/ideas taught by Deep Learning Systems - Algorithms and Implementation in CMU.

Feature Roadmap

• Tensor Impl and required operators
• Calculations in Operators (forward compute & backward gradient)
• Two-layer MLP with Back Propagation
• Reversed Automatic Differentiation
• Training / Testing Loops
• [] Normalization

Development

Code Format and Linter

All the coding conventions are mostly ensured by pre-commit, following

python -m pip install pre-commit
pre-commit install
# pre-commit will act with .pre-commit-config.yaml
pre-commit run

Environment Setup

This project and its dependencies is managed by uv, run the scripts below to setup the working environment:

uv init

Tests

Core functions are covered by Unit Tests

uv run pytest

uv run pytest test/test_autograd.py
uv run pytest test/test_simple_ml.py

uv run pytest -k "compute_gradient"

Integration test could be launched via

uv run python grady/app/simple_ml.py

Acknowlegement

This project is rooted and optimized from Needle and all outstanding scaffold from CMU Deep Learning System Course.