Saved in:
| Main Authors: | , , , , , , , , , , , , , , , , , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2412.04468 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866915956477919232 |
|---|---|
| author | Liu, Zhijian Zhu, Ligeng Shi, Baifeng Zhang, Zhuoyang Lou, Yuming Yang, Shang Xi, Haocheng Cao, Shiyi Gu, Yuxian Li, Dacheng Li, Xiuyu Fang, Yunhao Chen, Yukang Hsieh, Cheng-Yu Huang, De-An Cheng, An-Chieh Nath, Vishwesh Hu, Jinyi Liu, Sifei Krishna, Ranjay Xu, Daguang Wang, Xiaolong Molchanov, Pavlo Kautz, Jan Yin, Hongxu Han, Song Lu, Yao |
| author_facet | Liu, Zhijian Zhu, Ligeng Shi, Baifeng Zhang, Zhuoyang Lou, Yuming Yang, Shang Xi, Haocheng Cao, Shiyi Gu, Yuxian Li, Dacheng Li, Xiuyu Fang, Yunhao Chen, Yukang Hsieh, Cheng-Yu Huang, De-An Cheng, An-Chieh Nath, Vishwesh Hu, Jinyi Liu, Sifei Krishna, Ranjay Xu, Daguang Wang, Xiaolong Molchanov, Pavlo Kautz, Jan Yin, Hongxu Han, Song Lu, Yao |
| contents | Visual language models (VLMs) have made significant advances in accuracy in recent years. However, their efficiency has received much less attention. This paper introduces NVILA, a family of open VLMs designed to jointly optimize efficiency and accuracy. Building on top of VILA, we improve its model architecture by first scaling up the spatial and temporal resolutions, and then compressing visual tokens. This "scale-then-compress" approach enables NVILA to efficiently process high-resolution images and long videos. We further conduct a systematic investigation that enhances NVILA's efficiency throughout its entire lifecycle, from training and fine-tuning to deployment. NVILA matches or surpasses the accuracy of leading open and proprietary VLMs across a wide range of image and video benchmarks. At the same time, it reduces training cost by 1.9-5.1x, prefilling latency by 1.6-2.2x, and decoding latency by 1.2-2.8x. We release our code and models to facilitate reproducibility. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2412_04468 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | NVILA: Efficient Frontier Visual Language Models Liu, Zhijian Zhu, Ligeng Shi, Baifeng Zhang, Zhuoyang Lou, Yuming Yang, Shang Xi, Haocheng Cao, Shiyi Gu, Yuxian Li, Dacheng Li, Xiuyu Fang, Yunhao Chen, Yukang Hsieh, Cheng-Yu Huang, De-An Cheng, An-Chieh Nath, Vishwesh Hu, Jinyi Liu, Sifei Krishna, Ranjay Xu, Daguang Wang, Xiaolong Molchanov, Pavlo Kautz, Jan Yin, Hongxu Han, Song Lu, Yao Computer Vision and Pattern Recognition Visual language models (VLMs) have made significant advances in accuracy in recent years. However, their efficiency has received much less attention. This paper introduces NVILA, a family of open VLMs designed to jointly optimize efficiency and accuracy. Building on top of VILA, we improve its model architecture by first scaling up the spatial and temporal resolutions, and then compressing visual tokens. This "scale-then-compress" approach enables NVILA to efficiently process high-resolution images and long videos. We further conduct a systematic investigation that enhances NVILA's efficiency throughout its entire lifecycle, from training and fine-tuning to deployment. NVILA matches or surpasses the accuracy of leading open and proprietary VLMs across a wide range of image and video benchmarks. At the same time, it reduces training cost by 1.9-5.1x, prefilling latency by 1.6-2.2x, and decoding latency by 1.2-2.8x. We release our code and models to facilitate reproducibility. |
| title | NVILA: Efficient Frontier Visual Language Models |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2412.04468 |