Skip to content

Latest commit

 

History

History
123 lines (93 loc) · 6.58 KB

INSTALL.md

File metadata and controls

123 lines (93 loc) · 6.58 KB
Raw

🛠️Installation：

Basic requirements

  • Linux
  • Python 3.8+
  • PyTorch 1.12+
  • CUDA 11.6+
  • GCC & G++ 5.4+
  • GPU Memory >= 17G for loading basic tools(HuskyVQA, SegmentAnything, ImageOCRRecognition)

Install anaconda

Reference Installation Tutorial : English Tutorial or Chinese Tutorial

Clone project

Execute the following command in the root directory:

git clone https://github.com/OpenGVLab/InternGPT.git

Common errors and their resolutions :

  1. git installation status: done Execute the following command in the terminal git --version If it is not installed, execute the following command. sudo apt install git
  2. If the download speed is too slow: modify the preceding command as follows: git clone git://github.com/OpenGVLab/InternGPT.git

🗃Model zoo

our model_zoo has been released in huggingface! You can download it and directly place it into the root directory of this repo before running the app.

HuskyVQA, a strong VQA model, is also available in model_zoo. More details can refer to our report.

Note for husky checkpoint

Due to the license issuse, we could directly provide the checkpoint of Husky. The model_zoo contains the delta checkpoint between Husky and LLAMA.

To build the actual checkpoint of Husky, you need the original checkpoint of LLAMA, which should be put in model_zoo/llama/7B. We support automatically download the llama checkpoint, but you need to request a form for the download url from Meta (see here). Once you have the download url, paste it into PRESIGNED_URL="" at third-party/llama_download.sh.

Then, rerun the app would automatically download the original checkpoint, convert it to huggingface format, and build the Husky checkpoint.

Please make sure these folder model_zoo/llama, and model_zoo/llama_7B_hf contain the correct checkpoint, otherwise you should delete the folder and let the app download it again. Otherwise, you might encounter issuses similar as issue #5

model_zoo\llama_7B_hf does not appear to have a file named config.json.

FileNotFoundError: [Errno 2] No such file or directory: 'model_zoo/llama/7B/params.json'

Current project directory structure

tree

⛺Install dependencies

  • Create a virtual environment conda activate -n igpt python=3.8 1 2 1 3

  • Activate the virtual environment conda activate igpt Verify if you have successfully entered the virtual environment Inked1 4

  • Install essential dependencies conda install pytorch==1.13.0 torchvision==0.14.0 torchaudio==0.13.0 pytorch-cuda=11.6 -c pytorch -c nvidia notices:Download all dependencies simultaneously and ensure to include "-c pytorch -c nvidia" to avoid version conflicts. 1 5 1 6

  • Install additional dependencies pip install -r requirements.txt Inked1 7 1 8

  • Install detectron2

    • Install gcc & g++ apt-get install gcc apt-get install g++ apt-get install make notices:For most systems, the latest version of GCC and G++ will be downloaded automatically. To check the version of GCC, execute the command gcc --version. If the version is lower than 5.4, please refer to the following instructions to download the specific version: down gcc

    • Install detectron2 git clone https://github.com/facebookresearch/detectron2.git Inked1 9

      python -m pip install -e detectron2 Inked2 0 2 1

    • notices When encountering the messages "Compile without GPU support" or "Detectron2 CUDA compiler: Not available," please consider the following:

      python -c "import torch
from torch.utils.cpp_extension import CUDA_HOME
print(torch.cuda.is_available(), CUDA_HOME)"

Start the service

python -u app.py \ --load "ImageOCRRecognition_cuda:0,Text2Image_cuda:0,SegmentAnything_cuda:0,ActionRecognition_cuda:0,VideoCaption_cuda:0,DenseCaption_cuda:0,ReplaceMaskedAnything_cuda:0,LDMInpainting_cuda:0,SegText2Image_cuda:0,ScribbleText2Image_cuda:0,Image2Scribble_cuda:0,Image2Canny_cuda:0,CannyText2Image_cuda:0,StyleGAN_cuda:0,Anything2Image_cuda:0,HuskyVQA_cuda:0" -e -p 3456 --https

Now, you can access iGPT demo by visiting https://{ip}:3456 through your browser: result

Deploying with docker

  • Installation Please refer to the official documentation to install: Docker, Docker compose , and NVIDIA Container Toolkit.

  • Build and run an image Please add model_zoo and certificate folders to the root directory of this repo, and change /path/to/model_zoo and /path/to/certificate in docker/docker-compose.yml to model_zoo and certificate directories on your machine respectively.

    For more features of our iGPT, You can modify the load variable in the command section in the docker compose file.

    cd docker
# Build and run an image (require GPUs):
docker compose up