We study approximation algorithms for hypergraph partitioning objectives. We introduce polymatroidal cut functions amenable to metric embedding techniques, demonstrating an O(√log n)-approximation via metric relaxations. We also generalize the cut-matching game framework to achieve the first nearly-linear time O(log n)-approximation for standard hypergraph partitioning variants.