KFP SDK v2、v2 命名空间和 v2 兼容模式之间有什么区别？

Question

What is the difference between KFP SDK v2, the v2 namespace in KFP SDK v1, and v2 compatible mode?

Answer 1

The KFP SDK has two major versions: v1.8.x and v2.xx (in pre-release at the time of writing this).

KFP SDK v2.xx compiles pipelines and components to IR YAML [ example ], a platform neutral pipeline representation format. It can be run on the KFP open source backend or on other platforms, such as Google Cloud Vertex AI Pipelines .

KFP SDK v1.8.x , by default, compiles pipelines and components to Argo Workflow YAML . Argo Workflow YAML is executed on Kube.netes and is not platform neutral .

KFP SDK v1.8.x provides two ways to author pipelines using v2 Python syntax:

KFP SDK v2-compatible mode is a feature in KFP SDK v1.8.x which permits using v2 Python authoring syntax within KFP SDK v1 but compiles to Argo Workflow YAML. v2-compatible mode is deprecated and should not be used.

The KFP SDK v2 namespace in KFP SDK v1.8.x ( from kfp.v2 import dsl, compiler ) permits using v2 Python authoring syntax within KFP SDK v1 and compiles to IR YAML [ usage example ]. While this mode is not deprecated, users should prefer authoring IR YAML via the pre-released KFP SDK v2.xx .