VLM in a flash: I/O-Efficient Sparsification of Vision-Language Model via Neuron Chunking
By: Kichang Yang , Seonjun Kim , Minjae Kim and more
Potential Business Impact:
Makes AI models run faster on small devices.
Edge deployment of large Vision-Language Models (VLMs) increasingly relies on flash-based weight offloading, where activation sparsification is used to reduce I/O overhead. However, conventional sparsification remains model-centric, selecting neurons solely by activation magnitude and neglecting how access patterns influence flash performance. We present Neuron Chunking, an I/O-efficient sparsification strategy that operates on chunks (i.e., groups of contiguous neurons in memory) and couples neuron importance with storage access cost. The method models I/O latency through a lightweight abstraction of access contiguity and selects chunks with high utility, defined as neuron importance normalized by estimated latency. By aligning sparsification decisions with the underlying storage behavior, Neuron Chunking improves I/O efficiency by up to 4.65x and 5.76x on Jetson Orin Nano and Jetson AGX Orin, respectively.
Similar Papers
Resting Neurons, Active Insights: Improving Input Sparsification for Large Language Models
Machine Learning (CS)
Makes big computer brains work better and faster.
LVLM_CSP: Accelerating Large Vision Language Models via Clustering, Scattering, and Pruning for Reasoning Segmentation
CV and Pattern Recognition
Makes AI understand pictures faster, using less power.
HyperVL: An Efficient and Dynamic Multimodal Large Language Model for Edge Devices
CV and Pattern Recognition
Makes smart AI work on your phone.