Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2411.10669 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866910700801097728 |
|---|---|
| author | Long, Jinqiang Dai, Yanqi Yang, Guoxing Lin, Hongpeng Fei, Nanyi Gao, Yizhao Lu, Zhiwu |
| author_facet | Long, Jinqiang Dai, Yanqi Yang, Guoxing Lin, Hongpeng Fei, Nanyi Gao, Yizhao Lu, Zhiwu |
| contents | As the research of Multimodal Large Language Models (MLLMs) becomes popular, an advancing MLLM model is typically required to handle various textual and visual tasks (e.g., VQA, Detection, OCR, and ChartQA) simultaneously for real-world applications. However, due to the significant differences in representation and distribution among data from various tasks, simply mixing data of all tasks together leads to the well-known``multi-task conflict" issue, resulting in performance degradation across various tasks. To address this issue, we propose Awaker2.5-VL, a Mixture of Experts~(MoE) architecture suitable for MLLM, which acquires the multi-task capabilities through multiple sparsely activated experts. To speed up the training and inference of Awaker2.5-VL, each expert in our model is devised as a low-rank adaptation (LoRA) structure. Extensive experiments on multiple latest benchmarks demonstrate the effectiveness of Awaker2.5-VL. The code and model weight are released in our Project Page: https://github.com/MetabrainAGI/Awaker. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2411_10669 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Awaker2.5-VL: Stably Scaling MLLMs with Parameter-Efficient Mixture of Experts Long, Jinqiang Dai, Yanqi Yang, Guoxing Lin, Hongpeng Fei, Nanyi Gao, Yizhao Lu, Zhiwu Computer Vision and Pattern Recognition As the research of Multimodal Large Language Models (MLLMs) becomes popular, an advancing MLLM model is typically required to handle various textual and visual tasks (e.g., VQA, Detection, OCR, and ChartQA) simultaneously for real-world applications. However, due to the significant differences in representation and distribution among data from various tasks, simply mixing data of all tasks together leads to the well-known``multi-task conflict" issue, resulting in performance degradation across various tasks. To address this issue, we propose Awaker2.5-VL, a Mixture of Experts~(MoE) architecture suitable for MLLM, which acquires the multi-task capabilities through multiple sparsely activated experts. To speed up the training and inference of Awaker2.5-VL, each expert in our model is devised as a low-rank adaptation (LoRA) structure. Extensive experiments on multiple latest benchmarks demonstrate the effectiveness of Awaker2.5-VL. The code and model weight are released in our Project Page: https://github.com/MetabrainAGI/Awaker. |
| title | Awaker2.5-VL: Stably Scaling MLLMs with Parameter-Efficient Mixture of Experts |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2411.10669 |