SPHINX-X: Scaling Data and Parameters for a Family of Multi-modal Large Language Models

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Liu, Dongyang, Zhang, Renrui, Qiu, Longtian, Huang, Siyuan, Lin, Weifeng, Zhao, Shitian, Geng, Shijie, Lin, Ziyi, Jin, Peng, Zhang, Kaipeng, Shao, Wenqi, Xu, Chao, He, Conghui, He, Junjun, Shao, Hao, Lu, Pan, Li, Hongsheng, Qiao, Yu, Gao, Peng
Format: Preprint
Veröffentlicht: 2024
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866915206800605184
author Liu, Dongyang
Zhang, Renrui
Qiu, Longtian
Huang, Siyuan
Lin, Weifeng
Zhao, Shitian
Geng, Shijie
Lin, Ziyi
Jin, Peng
Zhang, Kaipeng
Shao, Wenqi
Xu, Chao
He, Conghui
He, Junjun
Shao, Hao
Lu, Pan
Li, Hongsheng
Qiao, Yu
Gao, Peng
author_facet Liu, Dongyang
Zhang, Renrui
Qiu, Longtian
Huang, Siyuan
Lin, Weifeng
Zhao, Shitian
Geng, Shijie
Lin, Ziyi
Jin, Peng
Zhang, Kaipeng
Shao, Wenqi
Xu, Chao
He, Conghui
He, Junjun
Shao, Hao
Lu, Pan
Li, Hongsheng
Qiao, Yu
Gao, Peng
contents We propose SPHINX-X, an extensive Multimodality Large Language Model (MLLM) series developed upon SPHINX. To improve the architecture and training efficiency, we modify the SPHINX framework by removing redundant visual encoders, bypassing fully-padded sub-images with skip tokens, and simplifying multi-stage training into a one-stage all-in-one paradigm. To fully unleash the potential of MLLMs, we assemble a comprehensive multi-domain and multimodal dataset covering publicly available resources in language, vision, and vision-language tasks. We further enrich this collection with our curated OCR intensive and Set-of-Mark datasets, extending the diversity and generality. By training over different base LLMs including TinyLlama1.1B, InternLM2-7B, LLaMA2-13B, and Mixtral8x7B, we obtain a spectrum of MLLMs that vary in parameter size and multilingual capabilities. Comprehensive benchmarking reveals a strong correlation between the multi-modal performance with the data and parameter scales. Code and models are released at https://github.com/Alpha-VLLM/LLaMA2-Accessory
format Preprint
id arxiv_https___arxiv_org_abs_2402_05935
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle SPHINX-X: Scaling Data and Parameters for a Family of Multi-modal Large Language Models
Liu, Dongyang
Zhang, Renrui
Qiu, Longtian
Huang, Siyuan
Lin, Weifeng
Zhao, Shitian
Geng, Shijie
Lin, Ziyi
Jin, Peng
Zhang, Kaipeng
Shao, Wenqi
Xu, Chao
He, Conghui
He, Junjun
Shao, Hao
Lu, Pan
Li, Hongsheng
Qiao, Yu
Gao, Peng
Computer Vision and Pattern Recognition
Artificial Intelligence
Computation and Language
Machine Learning
We propose SPHINX-X, an extensive Multimodality Large Language Model (MLLM) series developed upon SPHINX. To improve the architecture and training efficiency, we modify the SPHINX framework by removing redundant visual encoders, bypassing fully-padded sub-images with skip tokens, and simplifying multi-stage training into a one-stage all-in-one paradigm. To fully unleash the potential of MLLMs, we assemble a comprehensive multi-domain and multimodal dataset covering publicly available resources in language, vision, and vision-language tasks. We further enrich this collection with our curated OCR intensive and Set-of-Mark datasets, extending the diversity and generality. By training over different base LLMs including TinyLlama1.1B, InternLM2-7B, LLaMA2-13B, and Mixtral8x7B, we obtain a spectrum of MLLMs that vary in parameter size and multilingual capabilities. Comprehensive benchmarking reveals a strong correlation between the multi-modal performance with the data and parameter scales. Code and models are released at https://github.com/Alpha-VLLM/LLaMA2-Accessory
title SPHINX-X: Scaling Data and Parameters for a Family of Multi-modal Large Language Models
topic Computer Vision and Pattern Recognition
Artificial Intelligence
Computation and Language
Machine Learning
url https://arxiv.org/abs/2402.05935