LongLLaVA: Scaling Multi-modal LLMs to 1000 Images Efficiently via a Hybrid Architecture

TL;DR

LongLLaVA introduces a hybrid architecture and training strategy that significantly enhances multi-modal large language models, enabling efficient processing of up to 1000 images with high performance and low resource consumption.

Contribution

The paper presents a novel hybrid architecture combining Mamba and Transformer blocks, along with new data construction and training methods for scalable multi-modal LLMs.

Findings

Processes nearly 1000 images on a single GPU

Achieves competitive benchmark results

Maintains high throughput and low memory usage

Abstract

Expanding the long-context capabilities of Multi-modal Large Language Models~(MLLMs) is critical for advancing video understanding and high-resolution image analysis. Achieving this requires systematic improvements in model architecture, data construction, and training strategies, particularly to address challenges such as performance degradation with increasing image counts and high computational costs. In this paper, we propose a hybrid architecture that integrates Mamba and Transformer blocks, introduce data construction methods that capture both temporal and spatial dependencies, and employ a progressive training strategy. Our released model, LongLLaVA (\textbf{Long}-Context \textbf{L}arge \textbf{L}anguage \textbf{a}nd \textbf{V}ision \textbf{A}ssistant), demonstrates an effective balance between efficiency and performance. LongLLaVA achieves competitive results across various benchmarks while maintaining high throughput and low memory consumption. Notably, it can process nearly one thousand images on a single A100 80GB GPU, underscoring its potential for a wide range of multi-modal applications.