Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Aman, Euhid, Carlin, Esteban, Pao, Hsing-Kuo, Beltrame, Giovanni, Sari, Ghaluh Indah Permata, Chen, Yie-Tarng
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2510.10560
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866918159211036672
author Aman, Euhid
Carlin, Esteban
Pao, Hsing-Kuo
Beltrame, Giovanni
Sari, Ghaluh Indah Permata
Chen, Yie-Tarng
author_facet Aman, Euhid
Carlin, Esteban
Pao, Hsing-Kuo
Beltrame, Giovanni
Sari, Ghaluh Indah Permata
Chen, Yie-Tarng
contents Cross-attention transformers and other multimodal vision-language models excel at grounding and generation; however, their extensive, full-precision backbones make it challenging to deploy them on edge devices. Memory-augmented architectures enhance the utilization of past context; however, most works rarely pair them with aggressive edge-oriented quantization. We introduce BitMar, a quantized multimodal transformer that proposes an external human-like episodic memory for effective image-text generation on hardware with limited resources. BitMar utilizes 1.58-bit encoders, one for text (BitNet-style) and one for vision (DiNOv2-based), to create compact embeddings that are combined and used to query a fixed-size key-value episodic memory. During vector retrieval, the BitNet decoder applies per-layer conditioning, which increases the contextual relevance of generated content. The decoder also employs attention sinks with a sliding-window mechanism to process long or streaming inputs under tight memory budgets. The combination of per-layer conditioning and sliding-window attention achieves a strong quality-speed trade-off, delivering competitive captioning and multimodal understanding at low latency with a small model footprint. These characteristics make BitMar well-suited for edge deployment.
format Preprint
id arxiv_https___arxiv_org_abs_2510_10560
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle BitMar: Low-Bit Multimodal Fusion with Episodic Memory for Edge Devices
Aman, Euhid
Carlin, Esteban
Pao, Hsing-Kuo
Beltrame, Giovanni
Sari, Ghaluh Indah Permata
Chen, Yie-Tarng
Computation and Language
Artificial Intelligence
Computer Vision and Pattern Recognition
68T50
I.2.7
Cross-attention transformers and other multimodal vision-language models excel at grounding and generation; however, their extensive, full-precision backbones make it challenging to deploy them on edge devices. Memory-augmented architectures enhance the utilization of past context; however, most works rarely pair them with aggressive edge-oriented quantization. We introduce BitMar, a quantized multimodal transformer that proposes an external human-like episodic memory for effective image-text generation on hardware with limited resources. BitMar utilizes 1.58-bit encoders, one for text (BitNet-style) and one for vision (DiNOv2-based), to create compact embeddings that are combined and used to query a fixed-size key-value episodic memory. During vector retrieval, the BitNet decoder applies per-layer conditioning, which increases the contextual relevance of generated content. The decoder also employs attention sinks with a sliding-window mechanism to process long or streaming inputs under tight memory budgets. The combination of per-layer conditioning and sliding-window attention achieves a strong quality-speed trade-off, delivering competitive captioning and multimodal understanding at low latency with a small model footprint. These characteristics make BitMar well-suited for edge deployment.
title BitMar: Low-Bit Multimodal Fusion with Episodic Memory for Edge Devices
topic Computation and Language
Artificial Intelligence
Computer Vision and Pattern Recognition
68T50
I.2.7
url https://arxiv.org/abs/2510.10560