Explain Model Compression Techniques

Cover quantization-aware training, knowledge distillation, evaluation mode in deep learning frameworks, and contrastive learning. For each one, lay out its purpose, how it operates, when it helps, common pitfalls, and how you would judge whether it improved a production model.

Overview: This question probes understanding of model compression and representation-learning ideas—quantization-aware training, knowledge distillation, evaluation mode in deep learning frameworks, and contrastive learning—testing competence in model optimization, resource-aware deployment, and representation quality.