Sparse tensor contraction (SpTC) serves as an essential operation in high-performance applications. The high dimensionality of sparse tensors makes SpTC fundamentally challenging in aspects such as costly multi-dimensional index search, extensive intermediate output data and indirect addressing. Previous state-of-the-art work addresses some of these challenges through partial hash-table implementation. In this paper, we propose H3, a hash-table based and holistically optimized SpTC by providing a light-weighted sparse tensor oriented hash table design, proposing architecture-aware algorithm for hash table selection with size prediction, applying cross-stage optimizations to exploit shared information and avoid redundant operations. Evaluating on a set of tensors extracted from the real world, our H3 can achieve up to 8× (a geomean of 3×) speedup and reduce the memory footprint by up to 67% (an average of 50%) compared to the current state-of-the-art work.