Analyze model performance vs quality trade-offs across model compression/optimization techniques