Yuma-Ichikawa/qqa4co-bench

QQA4CO组合优化基准测试套件是一个统一的、预转换的、可直接用于基准测试的组合优化（CO）实例集合，适用于离散采样器、退火器和基于学习的求解器。该数据集包含了PQQA论文中引用的每个家族，以及一些广泛使用的社区基准测试（如G-set、DIMACS COLOR、Edwards-Anderson）。数据集设计为求解器无关，附带了一个Python加载器，但每个文件都是普通的`pickle(networkx.Graph)`或`numpy.savez`格式，因此可以从任何框架（如PyTorch、JAX、C++、Julia等）中使用。数据集包含八个配置名称（MaxCut、G-set、MIS、MaxClique、NormCut、Coloring、MIS-RRG、EA3D），涵盖了多种组合优化问题类型。

The QQA4CO Combinatorial Optimization Benchmark Suite is a unified, pre-converted, ready-to-benchmark collection of Combinatorial-Optimization (CO) instances for discrete samplers, annealers, and learning-based solvers. It includes every family referenced in the PQQA paper, along with several broadly used community benchmarks (G-set, DIMACS COLOR, Edwards-Anderson). The dataset is designed to be solver-agnostic, with a companion Python loader provided, but each file is in plain `pickle(networkx.Graph)` or `numpy.savez` format, making it usable from any framework (PyTorch, JAX, C++, Julia, etc.). The dataset consists of eight config_names (MaxCut, G-set, MIS, MaxClique, NormCut, Coloring, MIS-RRG, EA3D), covering a wide range of combinatorial optimization problems.