MSE Optimization Behavior Analysis in Sequential LLM Quantization

This study investigates how mean squared error (MSE) as a reconstruction loss function affects optimization during sequential post-training quantization (PTQ). It analyzes how optimization evolves across transformer blocks to provide a holistic view of learning-based quantization schemes.