PLM: Efficient Peripheral Language Models Hardware-Co-Designed for Ubiquitous Computing
Cheng Deng, Luoyang Sun, Jiwen Jiang, Yongcheng Zeng, Xinjian Wu, Wenxin Zhao, Qingfa Xiao, Jiachuan Wang, Haoyang Li, Lei Chen, Lionel M. Ni, Haifeng Zhang, Jun Wang
TL;DR
PLM tackles the challenge of deploying language models on edge devices by co-designing architecture with hardware constraints. It introduces Multi-head Latent Attention to reduce KV-cache memory and a squared ReLU activation to promote sparsity, coupled with a WSDC-based training regimen and ARIES-style RLHF for alignment. With a total pre-training corpus of approximately 2.48 trillion tokens and additional instruction-focused post-training data, PLM-1.8B demonstrates competitive performance on a broad suite of benchmarks while maintaining low activated-parameter counts and strong edge-deployment viability. The work also provides extensive analysis of attention mechanisms and sparsity, and outlines concrete future directions for peripheral LLMs, including on-device adaptation, privacy-preserving learning, and edge-cloud collaboration.
Abstract
While scaling laws have been continuously validated in large language models (LLMs) with increasing model parameters, the inherent tension between the inference demands of LLMs and the limited resources of edge devices poses a critical challenge to the development of edge intelligence. Recently, numerous small language models have emerged, aiming to distill the capabilities of LLMs into smaller footprints. However, these models often retain the fundamental architectural principles of their larger counterparts, still imposing considerable strain on the storage and bandwidth capacities of edge devices. In this paper, we introduce the PLM, a Peripheral Language Model, developed through a co-design process that jointly optimizes model architecture and edge system constraints. The PLM utilizes a Multi-head Latent Attention mechanism and employs the squared ReLU activation function to encourage sparsity, thereby reducing peak memory footprint during inference. During training, we collect and reorganize open-source datasets, implement a multi-phase training strategy, and empirically investigate the Warmup-Stable-Decay-Constant (WSDC) learning rate scheduler. Additionally, we incorporate Reinforcement Learning from Human Feedback (RLHF) by adopting the ARIES preference learning approach. Following a two-phase SFT process, this method yields performance gains of 2% in general tasks, 9% in the GSM8K task, and 11% in coding tasks. In addition to its novel architecture, evaluation results demonstrate that PLM outperforms existing small language models trained on publicly available data while maintaining the lowest number of activated parameters. Furthermore, deployment across various edge devices, including consumer-grade GPUs, mobile phones, and Raspberry Pis, validates PLM's suitability for peripheral applications. The PLM series models are publicly available at https://github.com/plm-team/PLM.