Domain
artificial intelligence, natural language processing, computer vision
Large language models (LLMs) have demonstrated impressive zero-shot abilities on a variety of open-ended tasks, while recent research has also explored the use of LLMs for multi-modal generation.In this study, we introduce mPLUG-Owl, a novel training paradigm that equips LLMs with multi-modal abilities through modularized learning of foundation LLM, a visual knowledge module, and a visual abstractor module.This approach can support multiple modalities and facilitate diverse unimodal and multimodal abilities through modality collaboration.The training paradigm of mPLUG-Owl involves a two-stage method for aligning image and text, which learns visual knowledge with the assistance of LLM, while maintaining and even improving the generation abilities of LLM.In the first stage, the visual knowledge module and abstractor module are trained with frozen LLM module to align the image and text.In the second stage, languageonly and multi-modal supervised datasets are used to jointly fine-tune a low-rank adaption (LoRA) module on LLM and the abstractor module by freezing the visual knowledge module.We carefully build a visually-related instruction evaluation set OwlEval.Experimental results show that our model outperform existing multi-modal models, demonstrating mPLUG-Owl's impressive instruction and visual understanding ability, multi-turn conversation ability and knowledge reasoning ability.Besides, we observe some unexpected and exciting abilities such as multi-image correlation and scene text understanding, which makes it possible to leverage it for harder real scenarios, such as vision-only document comprehension.Our code, pre-trained model, instruction-tuned models, and evaluation set are available at https://github.com/X-PLUG/mPLUG-Owl.The online demo is available at https://www.modelscope.cn/studios/damo/mPLUG-Owl.
This paper presents mPLUG-Owl, a novel training paradigm that enhances the multimodal capabilities of large language models (LLMs) through modularized learning of a foundational LLM, a visual knowledge module, and a visual abstractor module. The two-stage training method improves alignment between image and text and enhances the generation abilities of LLMs, effectively supporting multiple modalities and allowing for diverse unimodal and multimodal functionalities. The research introduces the OwlEval dataset for evaluating the capabilities of models in visual-related tasks and shows experimental results where mPLUG-Owl outperforms existing multimodal models in instruction understanding, knowledge reasoning, and multi-turn conversation abilities, showcasing unexpected skills in multi-image correlation and scene text understanding. The paper concludes by highlighting the practical applications and potential for further refinements in multimodal generation tasks.
This paper employs the following methods:
- Modularization
- Joint Instruction Tuning
- GPT-3
- BLOOM
- LLaMA
- GPT-4
- BLIP-2
- LLaVA
- MiniGPT-4
- OwlEval
- ViT-L/14
- LLaMA-7B
The following datasets were used in this research:
- LAION-400M
- COYO-700M
- Conceptual Captions
- MSCOCO
- LLaVA
- Instruction Understanding
- Visual Understanding
- Optical Character Recognition
- Knowledge Transfer Ability
- Reasoning Ability
- Multi-turn Dialogue Ability
- Outperforming existing multimodal models
- Improved instruction and visual understanding ability
- Enhanced multi-turn conversation and knowledge reasoning abilities
The authors identified the following limitations:
- Lack of multilingual training
- Limited performance in complex scene OCR
- Number of GPUs: None specified
- GPU Type: None specified
large language models
multimodal learning
modular training paradigm
visual knowledge module
visual abstractor