English

TOGGLE: Temporal Logic-Guided Large Language Model Compression for Edge

Artificial Intelligence 2025-12-19 v1 Logic in Computer Science

Abstract

Large Language Models (LLMs) deliver exceptional performance across natural language tasks but demand substantial computational resources, limiting their deployment on resource-constrained edge devices. Existing compression techniques, such as quantization and pruning, often degrade critical linguistic properties and lack formal guarantees for preserving model behavior. We propose Temporal Logic-Guided Large Language Model Compression (TOGGLE), a novel framework that leverages Signal Temporal Logic (STL) to formally specify and enforce linguistic properties during compression. TOGGLE employs an STL robustness-guided Bayesian optimization to systematically explore layer-wise quantization and pruning configurations, generating compressed models that formally satisfy specified linguistic constraints without retraining or fine-tuning. Evaluating TOGGLE on four LLM architectures (GPT-2, DeepSeek-V2 7B, LLaMA 3 8B, and Mistral 7B), we achieve up to 3.3x reduction in computational costs (FLOPs) and up to a 68.8% reduction in model size while satisfying all linguistic properties. TOGGLE represents the first integration of formal methods into LLM compression, enabling efficient, verifiable deployment of LLMs on edge hardware.

Keywords

Cite

@article{arxiv.2512.16855,
  title  = {TOGGLE: Temporal Logic-Guided Large Language Model Compression for Edge},
  author = {Khurram Khalil and Khaza Anuarul Hoque},
  journal= {arXiv preprint arXiv:2512.16855},
  year   = {2025}
}

Comments

Published in the IEEE ICCAD 2025 conference

R2 v1 2026-07-01T08:32:03.196Z