neural-fortran

Provides a clean, modular API that embeds seamlessly into existing Fortran codebases, eliminating the need for language bridging and reducing overhead.

Leverages OpenCoarrays for data-based parallelism, enabling efficient scaling of training and inference across multiple CPU cores, as detailed in the build instructions.

Through the nf-keras-hdf5 add-on, allows loading and deploying pre-trained dense and convolutional models from Keras HDF5 files, facilitating model reuse.

Supports diverse layers including embeddings, convolutions, and self-attention, enabling the construction of complex networks like transformers, as shown in the feature table.