Linux command
kfp 命令
网络
复制后可按需替换文件名、目录或参数。
常用示例
Create a pipeline run from a compiled pipeline
kfp run create -e [experiment] -r [run-name] -f [pipeline.yaml] --endpoint [http://localhost:8080]
List pipelines
kfp pipeline list --endpoint [http://localhost:8080]
Upload a pipeline
kfp pipeline create -p [pipeline-name] [pipeline.yaml]
Create an experiment
kfp experiment create -n [experiment-name] --endpoint [http://localhost:8080]
Compile a pipeline from Python
kfp dsl compile --py [pipeline.py] --output [pipeline.yaml]
Diagnose KFP installation
kfp diagnose_me
说明
kfp is the CLI for Kubeflow Pipelines, which orchestrates ML workflows as directed acyclic graphs (DAGs) on Kubernetes. Pipelines define reusable components with inputs, outputs, and dependencies. The CLI manages the full pipeline lifecycle: compiling Python pipeline definitions to YAML, uploading pipelines, creating experiments, and submitting runs. It connects to a running KFP backend via the `--endpoint` flag. Kubeflow itself is installed using Kustomize manifests via `kubectl apply -k` from the kubeflow/manifests repository. The older kfctl deployment tool is deprecated and archived.
参数
- run create|list|get|archive|unarchive|delete
- Manage pipeline runs.
- recurring-run create|list|get|enable|disable|delete
- Manage scheduled recurring runs.
- pipeline create|create-version|list|list-versions|get|delete
- Manage pipelines.
- experiment create|list|get|delete|archive|unarchive
- Manage experiments.
- dsl compile
- Compile a Python pipeline definition to YAML.
- component build
- Build a containerized component from a Python function.
- diagnose_me
- Run environment diagnostics (GCP-focused).
- -e, --experiment _name_
- Experiment name or ID.
- -r, --run-name _name_
- Name for the run.
- -f, --package-file _file_
- Compiled pipeline file (YAML).
- -p, --pipeline-name _name_
- Pipeline name.
- --endpoint _url_
- KFP API endpoint URL.
FAQ
What is the kfp command used for?
kfp is the CLI for Kubeflow Pipelines, which orchestrates ML workflows as directed acyclic graphs (DAGs) on Kubernetes. Pipelines define reusable components with inputs, outputs, and dependencies. The CLI manages the full pipeline lifecycle: compiling Python pipeline definitions to YAML, uploading pipelines, creating experiments, and submitting runs. It connects to a running KFP backend via the `--endpoint` flag. Kubeflow itself is installed using Kustomize manifests via `kubectl apply -k` from the kubeflow/manifests repository. The older kfctl deployment tool is deprecated and archived.
How do I run a basic kfp example?
Run `kfp run create -e [experiment] -r [run-name] -f [pipeline.yaml] --endpoint [http://localhost:8080]` in a terminal, then adjust file names, paths, flags, or remote targets for your system.
What does run create|list|get|archive|unarchive|delete do in kfp?
Manage pipeline runs.