Saved in:
Bibliographic Details
Main Authors: Roh, Seungwoo, Kim, Huiyeong, Kim, Jong-Chan
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2605.11678
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914557431119872
author Roh, Seungwoo
Kim, Huiyeong
Kim, Jong-Chan
author_facet Roh, Seungwoo
Kim, Huiyeong
Kim, Jong-Chan
contents End-to-end Vision-Language-Action (VLA) models for autonomous driving unify perception, reasoning, and control in a single neural network, achieving strong driving performance but requiring 20-60GB of GPU memory-far exceeding the 12-16GB available on commodity GPUs. We present a framework, which enables memory-efficient VLA inference on VRAM-constrained GPUs through system-level optimization alone, without model modification. Our work proceeds in three stages: (1) Sequential Demand Layering reduces VRAM usage from model-level to layer-level granularity; (2) Pipelined Demand Layering hides parameter transfer time within layer execution time via transfer--compute overlap; and (3) a GPU-Resident Layer Decision Policy, informed by per-module residency benefit analysis, eliminates the residual transfer overhead that pipelining cannot hide. We further propose a performance prediction model that determines the optimal configuration-both the number and placement of resident layers-from a single profiling run with less than 1.3% prediction error across all configurations. Applied to NVIDIA's Alpamayo-R1-10B (21.52GB) on an RTX 5070Ti (16GB), our work achieves up to 3.55x speedup over Accelerate offloading while maintaining full BF16 precision.
format Preprint
id arxiv_https___arxiv_org_abs_2605_11678
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle OOM-Free Alpamayo via CPU-GPU Memory Swapping for Vision-Language-Action Models
Roh, Seungwoo
Kim, Huiyeong
Kim, Jong-Chan
Artificial Intelligence
End-to-end Vision-Language-Action (VLA) models for autonomous driving unify perception, reasoning, and control in a single neural network, achieving strong driving performance but requiring 20-60GB of GPU memory-far exceeding the 12-16GB available on commodity GPUs. We present a framework, which enables memory-efficient VLA inference on VRAM-constrained GPUs through system-level optimization alone, without model modification. Our work proceeds in three stages: (1) Sequential Demand Layering reduces VRAM usage from model-level to layer-level granularity; (2) Pipelined Demand Layering hides parameter transfer time within layer execution time via transfer--compute overlap; and (3) a GPU-Resident Layer Decision Policy, informed by per-module residency benefit analysis, eliminates the residual transfer overhead that pipelining cannot hide. We further propose a performance prediction model that determines the optimal configuration-both the number and placement of resident layers-from a single profiling run with less than 1.3% prediction error across all configurations. Applied to NVIDIA's Alpamayo-R1-10B (21.52GB) on an RTX 5070Ti (16GB), our work achieves up to 3.55x speedup over Accelerate offloading while maintaining full BF16 precision.
title OOM-Free Alpamayo via CPU-GPU Memory Swapping for Vision-Language-Action Models
topic Artificial Intelligence
url https://arxiv.org/abs/2605.11678