LLaDA-V:具有视觉指令微调的大语言扩散模型
LLaDA-V: Large Language Diffusion Models with Visual Instruction Tuning
AI总结:
本文提出纯扩散多模态大语言模型 LLaDA-V,通过视觉编码器与MLP连接器实现视觉指令微调,在多模态理解上达到最先进性能,证明大语言扩散模型在多模态场景中的潜力。
摘要:
在这项工作中,我们介绍了 LLaDA-V,这是一个纯粹基于扩散的多模态大语言模型(MLLM),它将视觉指令微调与掩码扩散模型相结合,代表了与当前多模态方法中占主导地位的自回归范式的不同路径。LLaDA-V 建立在具有代表性的大语言扩散模型 LLaDA 之上,引入了一个视觉编码器和 MLP 连接器,将视觉特征投影到语言嵌入空间中,从而实现有效的多模态对齐。我们的实证研究揭示了几个有趣的结果:首先,尽管 LLaDA-V 的语言模型在纯文本任务上弱于 LLaMA3-8B 和 Qwen2-7B 等对应模型,但它仍展现出有前景的多模态性能。当在相同的指令数据上训练时,LLaDA-V 在多模态任务上与 LLaMA3-V 具有很强的竞争力,并且具有更好的数据可扩展性。它还缩小了与 Qwen2-VL 的性能差距,表明其架构在多模态任务上的有效性。其次,与现有的混合自回归-扩散模型和纯扩散多模态大语言模型相比,LLaDA-V 在多模态理解方面达到了最先进的性能。我们的研究结果表明,大语言扩散模型在多模态场景中展现出潜力,值得在未来的研究中进一步探索。项目页面和代码:https://ml-gsai.github.io/LLaDA-V-demo/。
英文摘要:
In this work, we introduce LLaDA-V, a purely diffusion-based Multimodal Large Language Model (MLLM) that integrates visual instruction tuning with masked diffusion models, representing a departure from the autoregressive paradigms dominant in current multimodal approaches. Built upon LLaDA, a representative large language diffusion model, LLaDA-V incorporates a vision encoder and MLP connector that projects visual features into the language embedding space, enabling effective multimodal alignment. Our empirical investigation reveals several intriguing results: First, LLaDA-V demonstrates promising multimodal performance despite its language model being weaker on purely textual tasks than counterparts like LLaMA3-8B and Qwen2-7B. When trained on the same instruction data, LLaDA-V is highly competitive to LLaMA3-V across multimodal tasks with better data scalability. It also narrows the performance gap to Qwen2-VL, suggesting the effectiveness of its architecture for multimodal tasks. Second, LLaDA-V achieves state-of-the-art performance in multimodal understanding compared to existing hybrid autoregressive-diffusion and purely diffusion-based MLLMs. Our findings suggest that large language diffusion models show promise in multimodal contexts and warrant further investigation in future research. Project page and codes: https://ml-gsai.github.io/LLaDA-V-demo/.
机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。