Flexible residual connections for PyTorch with a clean API. Build complex skip connections using Record/Apply modules. 5 operations (add, concat, gated, highway, multiply), learnable alphas, automatic projection. Thread-safe for DataParallel/DDP. Fully typed and tested.
python machine-learning deep-learning pytorch neural-networks densenet resnet mypy type-hints residual-learning model-building ml-library sequential-models skip-connections architecture-design pytorch-library neural-network-architecture residual-connections pytorch-modules highway-modules
-
Updated
Feb 18, 2026 - Python