Qm 𝗤𝘂𝗶𝗽 𝗡𝗲𝘁𝘄𝗼𝗿𝗸: 𝗢𝗻𝗲 𝗧𝗼𝗼𝗹𝗰𝗵𝗮𝗶𝗻, 𝗠𝘂𝗹𝘁𝗶𝗽𝗹𝗲 𝗖𝗼𝗺𝗽𝘂𝘁𝗲 𝗧𝗮𝗿𝗴𝗲𝘁𝘀 Different computing machines have different capabilities, but developers shouldn't have to rebuild their optimization problems for every hardware target. That's where XQuad comes in. 𝗔 𝗖𝗼𝗺𝗺𝗼𝗻 𝗟𝗮𝗻𝗴𝘂𝗮𝗴𝗲 𝗙𝗼𝗿 𝗢𝗽𝘁𝗶𝗺𝗶𝘇𝗮𝘁𝗶𝗼𝗻 Think of XQuad like LLVM, but designed for optimization problems. It takes QUBO, Ising, and discrete models, then compiles them into bytecode for XQVM, Quip's virtual machine. 𝗖𝗵𝗼𝗼𝘀𝗲 𝗪𝗵𝗲𝗿𝗲 𝗧𝗵𝗲 𝗣𝗿𝗼𝗯𝗹𝗲𝗺 𝗥𝘂𝗻𝘀 Once compiled, the workload can target different computing environments: → D-Wave quantum processors → GPU annealers → Local CPU samplers → @quipnetwork 𝗕𝘂𝗶𝗹𝗱𝗶𝗻𝗴 𝗔𝗿𝗼𝘂𝗻𝗱 𝗥𝗲𝗮𝗹 𝗢𝗽𝘁𝗶𝗺𝗶𝘇𝗮𝘁𝗶𝗼𝗻 𝗧𝗮𝘀𝗸𝘀 XQuad focuses on problems such as traveling salesman, graph coloring, knapsack, set cover, maximum independent set, and portfolio optimization. 𝗙𝗿𝗼𝗺 𝗖𝗼𝗱𝗲 𝗧𝗼 𝗦𝗼𝗹𝘃𝗲𝗿 Developers can write constraint-programming DSL or XQASM assembly, compile it into an ".xqb" file, and submit it to a solver. The public release is still early, so changes are expected before version 1.0. What interests me is how XQuad separates the optimization problem from the hardware executing it. That could make experimenting with different computing targets much easier as Quip's ecosystem develops.