Skip to content

sweepx

The toolkit on top of the sweep engine: one package, pip install sweepx, imported one module at a time as sweepx.<module>.

Workflows

  • Tasks · sweepx.tasks

    The YAML task runner: FWI, LSRTM, RTM, forward and wavefield runs from one file, on one GPU or many. Command: sweep-tasks.

  • Agent · sweepx.agent

    Plain-language control: a local LLM turns a request into a validated run. Command: sweep-agent.

Libraries

  • IO · sweepx.io

    SEG-Y to gathers: header index, data plans, prefetched reads, models, acquisition geometry.

  • Loss · sweepx.loss

    Misfit functions for inversion, from L2 to envelope, phase and optimal transport.

  • NN · sweepx.nn

    Implicit FWI, the model as a network: SIREN, hash encoding, priors.

  • Opt · sweepx.opt

    FWI optimizers: L-BFGS, truncated Newton / Gauss-Newton, Adam, Shampoo / SOAP, preconditioners.

How the modules depend on each other

workflows   sweepx.agent  →  sweepx.tasks
libraries   sweepx.io    sweepx.loss    sweepx.nn    sweepx.opt
engine      sweep  (pip name: sweep-solver)

A module imports only from the layers below it: the agent drives the task runner, which builds on the engine and on sweepx.io, sweepx.loss and sweepx.nn. The libraries import neither each other nor the engine, so sweepx.loss, sweepx.nn or sweepx.opt drop into any PyTorch inversion loop. CI checks these rules with import-linter.