Multi-target Build

envd supports building different container images from one single file, for different purpose. Typical scenario includes:

• Use CPU for model development, need GPU for large scale training jobs
• Use envd in daily development, but want to containerize dependency and publish it for model serving
• Require both X86 and ARM platforms.

Usage

Command envd up -f {build_file}:{build_func} can specify the build target, by running the build_func in build_file.

For example, you can declare multiple functions in one envd file, following python's syntax

def build():
    base(dev=True)
    install.conda()
    install.python()
    install.vscode_extensions([
        "ms-python.python",
    ])

    install.python_packages([
        "tensorflow",
        "numpy",
    ])
    shell("zsh")
    config.jupyter(token="", port=8888)

def build_gpu():
    build() # include all dependency declared in build function
    install.cuda(version="11.2.2", cudnn="8")

Then, to start with cpu, any of commands below works the same way

• envd up, this will run build function in build.envd file in current working directory
• envd up -f :build, explicitly specify build function
• envd up -f build.envd, explicitly specify build.envd
• envd up -f build.envd:build, explicitly specify both build.envd and build function

Similarly, if you want to start with cuda support declared in build_gpu function, you can try

• envd up -f :build_gpu, explicitly specify build_gpu function
• envd up -f build.envd:build_gpu, explicitly specify both build.envd and build_gpu function