QuIDE: Mastering the Quantized Intelligence Trade-off via Active Optimization
Provides a reproducible evaluation protocol and fitness function for mixed-precision search, addressing the lack of a unified efficiency metric for quantized models.
QuIDE introduces a unified metric (Intelligence Index) that collapses compression-accuracy-latency trade-offs into a single score for quantized neural networks. Experiments across multiple settings reveal task-dependent Pareto knees (e.g., 4-bit optimal for MNIST/LLMs, 8-bit for complex CNNs).
There is currently no unified metric for evaluating the efficiency of quantized neural networks. We propose QuIDE, built around the Intelligence Index I = (C x P)/log_2(T+1), which collapses the compression-accuracy-latency trade-off into a single score. Experiments across six settings -- SimpleCNN (MNIST, CIFAR), ResNet-18 (ImageNet-1K), and Llama-3-8B -- show a task-dependent Pareto Knee. 4-bit quantization is optimal for MNIST and large LLMs, while 8-bit is the sweet spot for complex CNN tasks (ResNet-18 on ImageNet), where 4-bit PTQ collapses accuracy catastrophically. The accuracy-gated variant I' correctly flags these non-viable configurations that the raw I would reward. QuIDE provides a reproducible evaluation protocol and a ready-to-use fitness function for mixed-precision search.