GravOpt – 20k-node MAX-CUT in ~7 minutes on a single CPU core

pip install numpy numba # once python gravopt.py benchmarks/G81.edges 2000 ``bash pip install numpy numba # once python gravopt.py benchmarks/G81.edges 2000 Example output on G81 (20 000 nodes, 40 000 edges): textGraph: 20000 nodes, 40000 edges Cut value: 14703.00 Ratio: 0.367650 Time: 48466.29 sec total (~13 h full run) → but 99 % of quality already reached at ~1200 steps Early stopping = the real killer feature After ~1200 iterations on most graphs, improvement drops below 10⁻⁵ per step → stop early and get Tabu-level quality in minutes instead of hours. ``bash Benchmarks (Nov 2025) Benchmarks (Nov 2025) GraphNodesEdgesRatioTime for 99 % qualityFull run timeG8120 00040 0000.36765~6–8 minutes~13 hours More graphs in /benchmarks (G14, G22, G81 + random sparse instances). Features ~318 lines total (the whole algorithm is in gravopt.py) No external solvers, no GPU, no PyTorch/TF < 80 MB RAM even on 100k-node graphs Built-in early stopping (automatic after stagnation) Install Bashpip install numpy numba