Efficient Architectures

Small language models, depth-width tradeoffs, weight sharing, and architectures designed for a latency budget.