??‍♂️ KarmaVLM (相生)

? Introduction

KarmaVLM is a family of high efficiency and powerful visual language model (VLM) pretrained with interleaved image-text data at scale, enabling content comprehension, recognition, and multi-round conversations about images.

? News

• [2024/02] KarmaVLM is released.

⚡️Features

KarmaVLM offers the following features:

• High Efficiency: KarmaVLM focuses on exploring the capabilities of small parametric quantitative models on multimodal tasks. So, KarmaVLM can be efficiently deployed on most GPU cards and personal computers, and even on end devices such as mobile phones.

• Multi-round text-image conversations: KarmaVLM can take both text and images as inputs and produce text outputs. Currently, it supports multi-round visual question answering with one image.

• Strong image comprehension: KarmaVLM is adept at analyzing visuals, making it an efficient tool for tasks like extracting, organizing, and summarizing information from images.

?Model Zoo

Basically we have achieved SOTA among models of the same parameter order of magnitude, even beyond some models with larger parameters. More Benchmark evaluations are in progress!

?‍? Quick Start

Requirements and Installation

git clone https://github.com/X-D-Lab/KarmaVLM.git
cd KarmaVLM

conda create -n karmavlm python=3.10 -y
conda activate karmavlm

pip install --upgrade pip  # enable PEP 660 support
pip install modelscope
pip install -e .
pip install -e ".[train]"
pip install flash-attn --no-build-isolation

you can download model:X-D-Lab/KarmaVLM-Qwen1.5-0_5B,and vision tower:openai/clip-vit-large-patch14 by run download.py

python download.py

you need to change the path in the download.py to your path,also, you need to change the path of vision tower in the config.json to your local vision tower path

from modelscope import snapshot_download
import os
os.environ['CURL_CA_BUNDLE'] = ''
model_dir = snapshot_download('X-D-Lab/KarmaVLM-Qwen1.5-0_5B',cache_dir="your_path/KarmaVLM-Qwen1.5-0_5B/") #where you need change
model_dir = snapshot_download('thomas/clip-vit-large-patch14',cache_dir="your_path/clip-vit-large-patch14/")#where you need change

? Demo

1. CLI Inference python -m llava.serve.cli \ --model-path /path/to/karmavlm/model \ --image-file /path/to/the/test/image
2. Gradio Web UI

• Starting the Controller python -m llava.serve.gradio_web_server \ --controller http://localhost:10000 \ --model-list-mode reload --share ##(optional)
• Launching the Gradio Web Server python -m llava.serve.model_worker \ --host 0.0.0.0 \ --controller http://localhost:10000 \ --port 40000 \ --worker http://localhost:40000 \ --model-path /path/to/karmavlm/model

? License

This project utilizes certain datasets and checkpoints that are subject to their respective original licenses. Users must comply with all terms and conditions of these original licenses. The content of this project itself is licensed under the Apache license 2.0.

?‍ Architecture

We build our project based on LLaVA: Large Language and Vision Assistant.

? Contributors

? Acknowledgement

同时感谢上海人工智能实验室推出的书生·浦语大模型实战营、InternStudio，为我门的项目提供宝贵的技术指导和强大的算力支持!~

@misc{liu2023llava,
      title={Visual Instruction Tuning}, 
      author={Liu, Haotian and Li, Chunyuan and Wu, Qingyang and Lee, Yong Jae},
      publisher={NeurIPS},
      year={2023},
}

@article{qwen,
  title={Qwen Technical Report},
  author={Jinze Bai and Shuai Bai and Yunfei Chu and Zeyu Cui and Kai Dang and Xiaodong Deng and Yang Fan and Wenbin Ge and Yu Han and Fei Huang and Binyuan Hui and Luo Ji and Mei Li and Junyang Lin and Runji Lin and Dayiheng Liu and Gao Liu and Chengqiang Lu and Keming Lu and Jianxin Ma and Rui Men and Xingzhang Ren and Xuancheng Ren and Chuanqi Tan and Sinan Tan and Jianhong Tu and Peng Wang and Shijie Wang and Wei Wang and Shengguang Wu and Benfeng Xu and Jin Xu and An Yang and Hao Yang and Jian Yang and Shusheng Yang and Yang Yao and Bowen Yu and Hongyi Yuan and Zheng Yuan and Jianwei Zhang and Xingxuan Zhang and Yichang Zhang and Zhenru Zhang and Chang Zhou and Jingren Zhou and Xiaohuan Zhou and Tianhang Zhu},
  journal={arXiv preprint arXiv:2309.16609},
  year={2023}
}

@misc{2023internlm,
    title={InternLM: A Multilingual Language Model with Progressively Enhanced Capabilities},
    author={InternLM Team},
    howpublished = {\url{https://github.com/InternLM/InternLM-techreport}},
    year={2023}
}

? Star History