Repository navigation
v0.1.0 — pip install sweepx
pip install sweepx → import sweep.
SWEEP is now on PyPI as a single py3-none-any wheel that works with any PyTorch version and any Python 3 — the C/CUDA backend is JIT-compiled against your own torch on first use of impl='c', only for your GPU's architecture, then cached. No CUDA/torch/Python matrix, no version lock-in.
Install
pip install sweepx
python -c "import sweep; sweep.precompile()" # optional: build the CUDA backend now (~3-5 min, then cached)The compiled backend (impl='c') needs a CUDA GPU and nvcc >= 12.4 (set SWEEP_JIT_ALLOW_OLD_CUDA=1 to try 12.0–12.3). The pure-Python eager / JAX backends need no nvcc.
What's included
- Physics — Acoustic, Elastic, LSRTM, DAS, Vector Reflectivity, Anisotropic, all in 2D & 3D
- Memory — boundary saving + checkpointing, with async storage across GPU / host / disk
- Compute — CPU & GPU, PyTorch & JAX, high-performance C/CUDA backends
- AI — native PyTorch bindings to embed wave equations directly into neural networks
Distributions
sweep-solver0.1.0 — the engine (import assweep)sweepx0.1.0.post1 — the install handle (pip install sweepx)