Skip to content

roseNNa v0.1

Compare
Choose a tag to compare
@sbryngelson sbryngelson released this 18 Nov 14:02
· 43 commits to master since this release
c7bc6f8

This is the first public release of roseNNA. roseNNA is a fast, portable, and minimally-intrusive library for neural network inference. Its intended use case is large Fortran- and C-based HPC codebases. roseNNa currently supports RNNs, CNNs, and MLPs, though more architectures are in the works. The library is optimized Fortran and outperforms PyTorch (by a factor between 2 and 5x) for the relatively small neural networks used in physics applications, like CFD.