Tommy Li

@ckadner I tested with pipelines and MLX works fine on k8s 1.21. We can cut the 0.2 branch and cherry pick new changes as we need to. This should let...

I cut a new branch here. Any bugs fixes and version change can go to here https://github.com/machine-learning-exchange/mlx/tree/v0.2.0

i added this pr for v0.2.0 branch, main branch should keep using the nightly tag https://github.com/machine-learning-exchange/mlx/pull/309/files

@ckadner when you cut the release, you can cut it from the v0.2.0 branch so that we can keep track of the diff for any new cherry pick commits.

Thanks @shawnzhu, what openshift environment did you test this on. We were testing it with Fyre Ember OCP 4.3 for this deployment.

The problem is that the metric value only incremented during pipeline create and pipeline upload. We need to make sure the metrics is reflecting the accurate number. https://github.com/kubeflow/kfp-tekton/blob/9307b361fcc005c7fc7b2c7376426f8a5a4ad01d/backend/src/apiserver/server/pipeline_server.go#L143-L145

Looks like there's a bug with using newlines. This is fixed in the latest kfp code https://github.com/kubeflow/pipelines/pull/4993 For a quick fix, you can run `pip install kfp==v1.4.0-rc1`. We will also...

> > Looks like there's a bug with using newlines. This is fixed in the latest kfp code > > [kubeflow/pipelines#4993](https://github.com/kubeflow/pipelines/pull/4993) > > For a quick fix, you can run...

@fenglixa so we only need to enable big data passing when the user defines the output as outputPath or input as inputPath. https://github.com/kubeflow/pipelines/blob/master/sdk/python/kfp/components/_python_op.py#L78-L81

If we don't have to use big data passing, then we will default it back to Tekton results.