DP-LLM: Runtime Model Adaptation with Dynamic Layer-wise Precision Assignment

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Kwon, Sangwoo, Seo, Seong Hoon, Lee, Jae W., Park, Yeonhong
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911305839935488
author Kwon, Sangwoo
Seo, Seong Hoon
Lee, Jae W.
Park, Yeonhong
author_facet Kwon, Sangwoo
Seo, Seong Hoon
Lee, Jae W.
Park, Yeonhong
contents How can we effectively handle queries for on-device large language models (LLMs) with varying runtime constraints, such as latency and accuracy? Multi-scale quantization addresses this challenge by enabling memory-efficient runtime model adaptation of LLMs through the overlaying of multiple model variants quantized to different bitwidths. Meanwhile, an important question still remains open-ended: how can models be properly configured to match a target precision or latency? While mixed-precision offers a promising solution, we take this further by leveraging the key observation that the sensitivity of each layer dynamically changes across decoding steps. Building on this insight, we introduce DP-LLM, a novel mechanism that dynamically assigns precision to each layer based on input values. Experimental results across multiple models and benchmarks demonstrate that DP-LLM achieves a superior performance-latency trade-off, outperforming prior approaches.
format Preprint
id arxiv_https___arxiv_org_abs_2508_06041
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle DP-LLM: Runtime Model Adaptation with Dynamic Layer-wise Precision Assignment
Kwon, Sangwoo
Seo, Seong Hoon
Lee, Jae W.
Park, Yeonhong
Machine Learning
Artificial Intelligence
How can we effectively handle queries for on-device large language models (LLMs) with varying runtime constraints, such as latency and accuracy? Multi-scale quantization addresses this challenge by enabling memory-efficient runtime model adaptation of LLMs through the overlaying of multiple model variants quantized to different bitwidths. Meanwhile, an important question still remains open-ended: how can models be properly configured to match a target precision or latency? While mixed-precision offers a promising solution, we take this further by leveraging the key observation that the sensitivity of each layer dynamically changes across decoding steps. Building on this insight, we introduce DP-LLM, a novel mechanism that dynamically assigns precision to each layer based on input values. Experimental results across multiple models and benchmarks demonstrate that DP-LLM achieves a superior performance-latency trade-off, outperforming prior approaches.
title DP-LLM: Runtime Model Adaptation with Dynamic Layer-wise Precision Assignment
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2508.06041