Neural Implicit Flow — PyTorch v2.0.0

NIF provides mesh-agnostic dimensionality reduction for parametric spatiotemporal fields. A ParameterNet maps time or physical parameters to the weights of a coordinate-dependent ShapeNet, which evaluates the field at spatial query points.

Original Neural Implicit Flow workflow illustration

The native PyTorch package includes multiscale neural fields, linear spatial bases, differentiable Jacobians and Hessians, regularization, streamed data, and optimization tools. The examples cover training, checkpoints, distributed learning, monitoring, pruning, and quantized inference.

Original comparison of field representations

Start with Neural Implicit Flow (NIF): mesh-agnostic dimensionality reduction, then explore the eight scientific cases in Eight scientific tutorials. Python 3.10+ and PyTorch 2.6+ are required; CUDA is optional. The API pages below are generated from the actual implementation imported during the documentation build.

Getting started

Scientific definitions and validation

Release and historical references

The historical documentation is preserved with its original examples and assets. Use the PyTorch guides and API reference above for version 2.0.0.