sweepx

One-line install for the sweep wave-equation engine.

sweepx is a thin umbrella package with no code of its own — installing it pulls in the sweep engine so you can import sweep.

Install

pip install sweepx

This currently pulls in:

Growing umbrella. The remaining companion packages (sweep-io, sweep-nn, sweep-opt, sweep-loss, sweep-viz, sweep-preproc, sweep-tasks, sweep-runner) are not published yet. When they ship, they'll be added to sweepx's dependencies — you keep the same pip install sweepx.

Import

After pip install sweepx, import as sweep, not sweepx:

import sweep
from sweep.propagator.torch import PropTorch     # PyTorch propagator
from sweep import equations, propagator          # native submodules

Same pattern as pip install scikit-learn → import sklearn: the import name sweep was already taken on PyPI, so the installable is named sweepx.

CUDA backend

sweep's GPU backend (impl='c') is JIT-compiled against your own PyTorch on first use — so a single wheel works with any torch version and any Python 3, with no prebuilt CUDA/torch/Python matrix. It needs a CUDA GPU and nvcc >= 12.4 (a system toolkit, module load cuda, or conda install -c nvidia cuda-toolkit):

import sweep
sweep.precompile()   # optional: build the CUDA backend now (~3-5 min, then cached)

Drop precompile() and the compile happens automatically on first use of impl='c', cached thereafter in ~/.cache/torch_extensions. The pure-Python eager / JAX backends need no nvcc.

Check what's available without triggering a compile:

import sweep
print(sweep.is_torch_binding_available())              # torch + CUDA GPU + nvcc present?
print(sweep.backend.torch.binding.diagnostics())       # usable / reason / cuda_home / built

License

MIT — see LICENSE.