Using Custom Base Image

By default, Tungstenkit uses a Python or CUDA image as the base image for containerization. But, you can manually set a base image.

You can use a custom base image by setting base_image in the define_model decorator:

@define_model(
    input=SDInput,
    output=SDOutput,
    batch_size=1,
    gpu=True,
    gpu_mem_gb=14,
    base_image="mjpyeon/tungsten-sd-base:v1",
)
class StableDiffusion:
    ...

In this case, since Tungstenkit doesn't know which OS the base image contains, system_packages argument of define_model decorator will be ignored.