Decorators#
container_spec#
from kfp_toolbox import container_spec
container_spec decorator specifies computing resources to be used by the component.
To apply this to a Python function-based component, it must be added outside of the component decorator.
from kfp.v2 import dsl
@container_spec(cpu="2", memory="16G")
@dsl.component()
def component_function():
...
For other components, wrap the component as a function.
from kfp.components import load_component_from_file
component = load_component_from_file("path/to/component.yaml")
component = container_spec(cpu="2", memory="16G")(component)
If multiple container_spec decorators are stacked, the one placed further out will take precedence. For example, suppose you have created an alias default_spec. If you want to overwrite part of it, place a new container_spec decorator outside of the default_spec decorator to overwrite it.
from kfp.v2 import dsl
default_spec = container_spec(cpu="2", memory="16G")
@container_spec(cpu="1")
@default_spec
@dsl.component()
def component_function():
...
See all available options here:
option |
type |
description |
examples |
|---|---|---|---|
cpu |
str |
CPU limit |
|
memory |
str |
Memory limit |
|
gpu |
str |
GPU limit |
|
accelerator |
str |
Accelerator type |
|