Seu1ki/SMOOTH: [ISCA 2026] Artifact for "SMOOTH: Hardware-Assisted Fine-Grained On-Chip Memory Management for Efficient On-Device LLM Inference"

Read full text Summarise with VIRAT
Authors: Seu1ki
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Artifact for "SMOOTH: Hardware-Assisted Fine-Grained On-Chip Memory Management for Efficient On-Device LLM Inference" This release contains the official simulation artifacts and architectural models for our paper accepted at the 53rd International Symposium on Computer Architecture (ISCA 2026). This repository provides the necessary source code, configurations, and baseline implementations to evaluate the SMOOTH memory management framework and reproduce the key experiments. Paper Information Title: SMOOTH: Hardware-Assisted Fine-Grained On-Chip Memory Management for Efficient On-Device LLM Inference Conference: The 53rd Annual International Symposium on Computer Architecture (ISCA 2026) What's in this Release? This artifact is built on top of LLMCompass (an extension of ScaleSim) and includes hardware synthesis scripts. The repository is structured across multiple branches to facilitate comparison against various baseline strategies: main: SMOOTH-ER (Full version with block allocation + early reclamation) smooth: SMOOTH-Base (Block allocation only) compiler-ideal: Ideal compiler-based allocation (best-fit + full preloading) gemmini: Pipelined allocation with tile overlap capuchin: Cache-like hardware-managed strategy Quick Start & Documentation For comprehensive instructions on installation, dependencies, and running the cycle-accurate simulator, please refer to the main documentation: Read the README.md in the repository for detailed steps.

Fields of Study & Subjects

Parallel Computing and Optimization Techniques Embedded Systems Design Techniques Low-power high-performance VLSI design

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
30 March 2026
Language
EN