Skip to content

Latest commit

 

History

History
77 lines (63 loc) · 2.78 KB

install.md

File metadata and controls

77 lines (63 loc) · 2.78 KB
Raw

Install

The install.py file creates a venv for the project, installs the required packages and symlinks mlapi.py to /usr/local/bin/mlapi. Pytorch and onnxruntime are installed by default, face-recognition with D-Lib and opencv are additional (--face-recognition and --opencv).

Requirements

  • Clone this repo and cd into the repo directory.
  • System packages (package names from debian based system)
    • python3-venv
    • python3-pip
    • python3-setuptools
    • python3-wheel
  • Python 3.8+ with packages
    • requests
    • psutil
    • distro
    • tqdm
  • --user and --group arguments are required
  • --cpu or --gpu argument is required

Caution

the pypi (pip) based pycoral libs only support python 3.7-3.9. If you are using python 3.10, you will need to install the pycoral libs manually, please see: this issue and read through the whole comment thread. You could also build the pycoral libs from source for python 3.11+ using the method and hacks in that issue thread.

Processor type

Important

The --cpu or --gpu <choice> argument is required.

--gpu choices:

  • cuda12.1
  • cuda11.8
  • rocm6.0

Note

The --cpu argument does not require a value, it is a flag to install the CPU based packages.

Install as user

Important

The --user and --group arguments are required.

These are the user and group that the server will run as.

OpenCV

  • --opencv installs the opencv-contrib-python package (no CUDA support) opencv is a requirement for zomi-server to run.

Tip

Advanced users can compile opencv with CUDA/cuDNN support and link it into the venv to use GPU acceleration in opencv. A doc will be written outlining the linking process, there are several build tutorials online.

Additional ML frameworks

  • --face-recognition installs the face-recognition package with D-Lib (may require building D-Lib first with CUDA for nvidia support)
  • Others are planned for the future (deepface, mmdetection, etc.)

Examples

CPU

python3 examples/install.py --cpu --debug --opencv --dry-run --user <username> --group <groupname>

Nvidia GPU using CUDA 12.1

python3 examples/install.py --gpu cuda12.1 --debug --opencv --dry-run --user <username> --group <groupname>

Nvidia GPU using CUDA 11.8

python3 examples/install.py --gpu cuda11.8 --debug --opencv --dry-run --user <username> --group <groupname>

AMD GPU using ROCm 6.0

python3 examples/install.py --gpu rocm6.0 --debug --opencv --dry-run --user <username> --group <groupname>

Tip

--dry-run will show the commands that will be run without actually running them.