MEGA Hub

PolyQ: Codesigning End-to-End Quantization Framework for Scalable Edge CPU LLM Inference

Authors

Do you know Hyunwoo Oh?You can claim authorship or link another user.Do you know Suyeon Jang?You can claim authorship or link another user.Do you know Hanning Chen?You can claim authorship or link another user.Do you know KyungIn Nam?You can claim authorship or link another user.Do you know Sanggeon Yun?You can claim authorship or link another user.Do you know Ryozo Masukawa?You can claim authorship or link another user.Do you know Mohsen Imani?You can claim authorship or link another user.

Abstract

CPUs are the most universal target for on-device LLM inference, but existing low-bit quantization methods offer either coarse operating points or fine-grained mixed precision that is difficult to execute efficiently on CPUs. We present PolyQ, a CPU-oriented compiler/quantization co-design for activation-aware channel-wise bit allocation under a user-specified average-bit budget. PolyQ assigns per-channel bit-widths from $\{2,3,4,8,16\}$, then uses a compile-time model compiler to permute and cluster channels into bit-homogeneous blocks, generate SIMD- and LUT-compatible kernels, and merge compatible permutations across operators to keep layout regularization off the runtime path. This turns fine-grained budget fitting into a practical fractional-bit deployment method for CPU-only inference. Across Falcon-H1-3B, Llama2-13B, and Qwen3-32B on WikiText-2, PolyQ provides stable quality scaling from 3--6\,b and improves perplexity by 2.4--32.1\% over prior methods at a 3\,b target. End-to-end measurements on three representative CPUs -- workstation, laptop, and mobile -- show that compiler layout regularization reduces activation reorder traffic by up to 70.8\%, prefill latency and decode throughput scale nearly proportionally with the configured bit budget, and energy/token overhead stays below 2\% relative to an optimized LUT-based back-end. These results show that fractional-bit CPU deployment is practical, predictable, and energy-efficient across diverse edge targets.

Community

00

Publication notes

Author note
Accepted to ICCAD 2026