{"ID":2834453,"CreatedAt":"2026-06-01T04:54:23.091178241Z","UpdatedAt":"2026-06-01T04:54:23.091178241Z","DeletedAt":null,"paper_url":"https://arxiv.org/abs/2512.01546","arxiv_id":"2512.01546","title":"LPCD: Unified Framework from Layer-Wise to Submodule Quantization","abstract":"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.","short_abstract":"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...","url_abs":"https://arxiv.org/abs/2512.01546","url_pdf":"https://arxiv.org/pdf/2512.01546v1","authors":"[\"Yuma Ichikawa\",\"Yudai Fujimoto\",\"Akira Sakai\"]","published":"2025-12-01T11:21:18Z","proceeding":"stat.ML","tasks":"[\"stat.ML\",\"cs.AI\",\"cs.CL\",\"cs.LG\"]","methods":"[\"Large Language Model\"]","has_code":false}