qlabs-eng / noise

Speedrun for pretraining without backpropagation

qlabs-eng / dust

Pretraining Transformers Without Backpropagation

HIPS / autograd

Efficiently computes derivatives of NumPy code.

dev-luigi / neural-network-digits

Neural network from scratch in pure Python + NumPy that recognises handwritten digits (MNIST), with an educational GUI: watch it learn, tweak it while it trains, look inside every layer.

hewanglanhao / deepvision-studio

Interactive deep learning lab for visualizing forward/backpropagation, training real PyTorch models, and exploring CNNs, Transformers, RNNs, and AI history.

Wrste / pheromone-network

A zero-dependency TypeScript memory kernel that gives any agent long-term memory. No training or backpropagation: each output reads a few tag-compatible neighbors, updated by a manual Hebbian rule. Links used often are reinforced; idle ones evaporate by half-life. Based on the paper Local Pheromone Network (arXiv:2606.30669).

karpathy / nn-zero-to-hero

A video course with Jupyter notebooks that builds neural networks and backpropagation from scratch, from basics to Transformer language models.