ANOLE: An Open, Autoregressive, Native Large Multimodal Models for Interleaved Image-Text Generation

TL;DR

Anole is an open-source, autoregressive large multimodal model capable of interleaved image and text generation, addressing limitations of previous models by integrating visual and textual modalities natively and efficiently.

Contribution

It introduces Anole, a native, autoregressive multimodal model built on Chameleon, with a novel fine-tuning strategy for efficient, high-quality image-text generation.

Findings

High-quality, coherent multimodal generation demonstrated

Open-sourced model and training framework provided

Efficient fine-tuning strategy enhances performance

Abstract

Previous open-source large multimodal models (LMMs) have faced several limitations: (1) they often lack native integration, requiring adapters to align visual representations with pre-trained large language models (LLMs); (2) many are restricted to single-modal generation; (3) while some support multimodal generation, they rely on separate diffusion models for visual modeling and generation. To mitigate these limitations, we present Anole, an open, autoregressive, native large multimodal model for interleaved image-text generation. We build Anole from Meta AI's Chameleon, adopting an innovative fine-tuning strategy that is both data-efficient and parameter-efficient. Anole demonstrates high-quality, coherent multimodal generation capabilities. We have open-sourced our model, training framework, and instruction tuning data.