Benchmarks

Artifex retains benchmark and evaluation modules, but the public benchmark surface is intentionally narrow. There is no supported public benchmark CLI runner, and several retained suites now require explicit demo_mode=True or synthetic/demo dataset settings before mock or placeholder paths are allowed to run.

Current Contract

• Use Python APIs for benchmark work; do not expect artifex benchmark to execute a supported runtime flow.
• Treat retained mock or synthetic suites as demo-only teaching surfaces.
• Expect supported-mode evaluation to fail fast when benchmark-grade assets or third-party dependencies are missing.

Retained Demo Walkthroughs

• Geometric Benchmark Demo
• Protein-Ligand Benchmark Demo
• Multi-β VAE Benchmark Demo

CLI Status

The CLI benchmark page is a retirement notice, not an execution guide: Benchmark CLI.

Explicit Demo Opt-In

from artifex.benchmarks.suites.multi_beta_vae_suite import MultiBetaVAEBenchmarkSuite

suite = MultiBetaVAEBenchmarkSuite(
    dataset_config={"num_samples": 100, "image_size": 64, "include_attributes": True},
    benchmark_config={"num_samples": 50, "batch_size": 10},
    demo_mode=True,
    rngs=rngs,
)

The same rule applies to retained synthetic dataset paths such as data_source="synthetic" and to mock metric backends such as mock_inception=True or mock_implementation=True: those switches are explicit demo-mode opt-ins, not the primary supported benchmark runtime.