LPCD: Unified Framework from Layer-Wise to Submodule Quantization
This work addresses the limitation of existing PTQ methods that focus only on layers, offering a more effective quantization approach for LLMs, though it is incremental as it builds on prior layer-wise and submodule techniques.
The paper tackles the problem of post-training quantization (PTQ) for large language models (LLMs) by introducing LPCD, a unified framework that extends PTQ beyond individual layers to optimize arbitrary submodules, resulting in consistent performance enhancements across diverse LLM architectures and bit-widths.
Post-training quantization (PTQ) aims to preserve model-level behavior; however, most methods focus on individual linear layers. Even recent extensions, such as QEP and LoaQ, which mitigate error propagation or target specific submodules, still rely on layer-wise formulations and fail to capture the behavior of larger submodules. We introduce Layer-Projected Coordinate Descent (LPCD), a unified framework that extends PTQ beyond layers by optimizing relaxed objectives across arbitrary submodules and projecting the solutions with layer-wise quantizers. LPCD generalizes existing methods and provides a principled approach to quantizing complex submodules while maintaining the efficiency and compatibility of layer-wise PTQ pipelines. Across diverse LLM architectures and bit-widths, LPCD-based submodule quantization consistently enhances both layer-wise PTQ methods and existing submodule approaches.