What are the trade-offs between different pruning strategies, structured vs unstructured, when deploying models to edge devices for TETRAMEM?
