A practical-scale reproduction of the 2014 sequence-to-sequence architecture, with explicit LSTM gates, controlled source-reversal experiments, and deliberately reported failure modes.
Technical highlightPairs an explicit NumPy LSTM forward pass and numerical gradient checks with a PyTorch-autograd training pipeline, then evaluates normal and reversed source order across four controlled runs.
- Python
- NumPy
- PyTorch
- LSTM
- NLP
- SacreBLEU
A NumPy-only character language model that implements the recurrent forward pass, BPTT, gradient checking, Adam, checkpointing, and controlled sampling by hand.
Technical highlightImplements and numerically verifies the complete recurrent training path without PyTorch, TensorFlow, or autograd, then records measured validation loss and generated text.
- Python
- NumPy
- RNN
- BPTT
- Adam
- Pytest
Learning extensions to Andrej Karpathy's micrograd, with input-shape fixes, gradient regression tests, a backpropagation walkthrough, and reproducible command-line training.
Technical highlightFixes silent input truncation and single-neuron hidden layers, then checks gradients against PyTorch and exercises a seeded training demo with automated tests.
- Python
- Autograd
- Backpropagation
- Pytest
- Graphviz
- GitHub Actions