Examples

This page lists runnable examples and recommended experiment entry points.

For the minimum usage pattern and the smallest inline optimizer example, start with Quick Start.

PILOT formal runs

The repository retains formal PILOT launcher scripts under workdirs/scripts/. These correspond to the preserved formal artifacts under workdirs/results/.

Install the PILOT/demo dependencies from the repository root:

cd INFTY
python -m pip install ".[examples]"

The minimal scripts under examples/infty_minimal/ do not require this extra.

Run the retained formal C-Flat + ESD job:

bash workdirs/scripts/run_memo_cflat_esd_full.sh

Run the retained formal ZeroFlow comparison suite:

bash workdirs/scripts/run_ease_zo_all_parallel.sh

Run the retained formal conflict-method suite, including UniGrad-FS:

bash workdirs/scripts/run_wa_conflicts_all_parallel.sh

Minimal examples included in this documentation package

This documentation package also contains small standalone scripts under:

examples/infty_minimal/

File	Purpose
minimal_cflat.py	One-step/small-loop C-Flat example on toy data.
minimal_zeroflow.py	ZeroFlow example using a finite-difference update.
minimal_unigrad_fs.py	Two-loss UniGrad-FS example.
minimal_visualization.py	Toy trajectory visualization example.

Run them from the repository root after installing INFTY:

python examples/infty_minimal/minimal_cflat.py
python examples/infty_minimal/minimal_zeroflow.py
python examples/infty_minimal/minimal_unigrad_fs.py
python examples/infty_minimal/minimal_visualization.py

Writing your own example

A good example should include:

1. environment assumptions;
2. dataset or toy-data creation;
3. base optimizer construction;
4. INFTY optimizer construction;
5. closure definition;
6. set_closure and step calls;
7. expected output or artifact path.

Example: loss-landscape visualization pattern

from infty import plot as infty_plot

result = infty_plot.visualize_landscape(
    optimizer=optimizer,
    model=model,
    create_loss_fn=create_loss_fn,
    loader=train_loader,
    task=task_id,
    device=device,
    output_dir="workdirs/plots/diagnostics/landscape/demo",
)
print(result["plot_path"])

Example: conflict visualization pattern

from infty import plot as infty_plot

result = infty_plot.visualize_conflicts(
    optimizer=optimizer,
    task=task_id,
    output_dir="workdirs/plots/diagnostics/conflicts/demo",
)
print(result)