Predict Before You Deploy: Offline Prediction of Quantization-Induced Task Degradation for World Action Models

A method to predict how quantization degrades task performance in world action models before deployment, improving reliability of compressed AI systems.

Auto-scored: Provides a real but partial technical step toward safer deployment of quantized models by predicting performance degradation offline.