ack-ai-pipeline is a workflow orchestration engine provided by the cloud-native AI suite. It is designed for job orchestration, experiment management, and model tracking. This component simplifies job management during AI development and improves the efficiency of model development and iteration. This topic introduces ack-ai-pipeline and describes the usage notes and release notes for ack-ai-pipeline.
Introduction
ack-ai-pipeline is developed based on the open source workflow framework Kubeflow Pipelines. Compared with Kubeflow Pipelines, ack-ai-pipeline is more stable, secure, and simple. You can use ack-ai-pipeline to build your own MLOps platform. ack-ai-pipeline provides the following capabilities.
Job orchestration: Use Kubernetes clusters for distributed Argo workflows to create various directed-acyclic graph (DAG) workflows. You can define, manage, and reuse workflows on demand.
Experiment management: Display and compare the results of workflows that use different experiment parameters, such as model hyperparameters. You can fine-tune the jobs in a workflow based on the comparison result.
Model tracking: Track the input and output of each step in a workflow. This helps you better troubleshoot and fine-tune models.
Usage notes
Only Container Service for Kubernetes (ACK) Pro clusters whose Kubernetes versions are 1.18 and later support ack-ai-pipeline. For more information about how to install and use ack-ai-pipeline, see Install and configure Kubeflow Pipelines and Use Kubeflow Pipelines to create workflows.
Description
May 2023
Version | Description | Release date | Impact |
1.0.3 | Security is enhanced. | 2023-05-11 | No impact on workloads |
April 2023
Version | Description | Release date | Impact |
1.0.2 | Security is enhanced. | 2023-04-17 | No impact on workloads |
March 2023
Version | Description | Release date | Impact |
1.0.1 | The OSS bucket name conflict issue is fixed. | 2023-03-14 | No impact on workloads |
July 2022
Version | Description | Release date | Impact |
1.0.0 | This is the first release. | 2022-07-04 | No impact on workloads |