IQP Generative Benchmark

This repository contains the scripts and cached artifacts for the IQP generative-model benchmark used in the manuscript figures. The active reporting standard is documented in STANDARD_TRAINING_PROTOCOL.md and docs/benchmark_reporting_protocol.md.

Active Protocol

• System size: n=12
• Target family: even-parity score-tilted distributions
• Beta sweep: 0.1, 0.2, ..., 2.0
• Matched seeds: 111..120
• Training samples per matched instance: m=200
• Full sweep size: 20 betas x 10 seeds = 200 matched instances
• Shared budget: 600 optimizer updates or epochs per model
• Reference parity band: sigma=1.0, K=512

The current full-sweep raw summary is:

Model	mean KL	median KL	KL wins	mean C_q(1000)
IQP-parity	0.385 +/- 0.021	0.414	190/200	0.053 +/- 0.004
Ising+fields (NN+NNN)	0.923 +/- 0.062	0.929	0/200	0.038 +/- 0.003
Dense Ising+fields	0.947 +/- 0.025	0.978	0/200	0.035 +/- 0.003
AR Transformer	0.744 +/- 0.054	0.737	10/200	0.036 +/- 0.002
MaxEnt-parity	1.804 +/- 0.108	1.689	0/200	0.018 +/- 0.002

Reproducibility

The final figures are regenerated from cached CSV/NPZ artifacts unless --recompute 1 is explicitly passed.

Common commands:

python experiment_2_beta_kl_summary.py \
  --outdir plots/experiment_2_beta_kl_summary \
  --series-csv plots/experiment_2_beta_kl_summary/experiment_2_beta_kl_summary_series.csv

python experiment_3_beta_quality_coverage.py \
  --outdir plots/experiment_3_beta_quality_coverage \
  --series-q1000 plots/experiment_3_beta_quality_coverage/experiment_3_beta_quality_coverage_q1000_series.csv

python make_aligned_kl_triptych.py
python make_aligned_recovery_fourpanel.py
python make_aligned_cross_class_diagnostics.py

Hardware sampling artifacts are already cached under plots/experiment_15_ibm_hardware_seedwise_best_coverage/. Re-running the hardware scripts requires configured IBM Quantum credentials and current backend availability.

Dependencies

Install the Python dependencies with:

python -m pip install -r requirements.txt

The checked-in requirements are pinned to the environment used for the current publication rerender pass (Python 3.13.2).

Exact LaTeX-sized figure rendering uses Matplotlib with text.usetex=True. A local TeX installation with newtx fonts is therefore required for exact paper rendering; cached PDFs are included for inspection without rerendering.