README.md

Benchmark

This benchmark suite compares the current development checkout against the repository main branch.

Run the full comparison from the repository root with:

git fetch origin main
julia --project=benchmark benchmark/benchmark.jl --run --main
julia --project=benchmark benchmark/benchmark.jl --run --dev
julia --project=benchmark benchmark/report.jl --run

The benchmark runner stores tuning parameters and benchmark results under benchmark/data/, and the report script writes benchmark/data/REPORT.md. The report compares main and dev using minimum-time estimates with a 15% tolerance to keep CI noise from showing up as false regressions. The reported times cover batched benchmark operations, so they should be read as relative comparison figures rather than per-call timings.

The constructors suite includes both dict-backed model construction and direct boolean MOI.ModelLike parsing through QUBOTools.Model(moi_model) so that performance work such as issue #56 is exercised by the benchmark harness on the reported QUBO conversion path.

It also includes a dense TSP-style constructor fixture (tsp/cities=36) that exercises the same public MOI extraction path used by ToQUBO integrations:

QUBOTools.Model(moi_model)

The default fixture is sized for pull-request benchmark runtime. To add the 10,000-variable TSP extraction case (tsp/cities=100) to the constructor suite, set QUBOTOOLS_SCALE_BENCHMARKS=true when running the benchmark commands.

To smoke-test the benchmark harness itself without running the full benchmark, use:

julia --project=benchmark benchmark/test/runtests.jl