duoan / TorchCode

Why you might care

Aimed at engineers preparing for ML interviews who need to write core operations like LayerNorm and multi-head attention from memory. Each problem has a judge that checks correctness, gradients, and timing, plus hints and reference solutions. Runs self-hosted via Docker or Podman, or on Hugging Face Spaces and Colab without setup.