While using the @kubernetes decorator on metaflow on a step, the memory and cpu represent the requests or limits ? If I have to specify the limits how do I specify them ?

To the best of my knowledge, explicit resource limits can't be directly configured using Metaflow's decorators; instead, those limits are controlled at the Kubernetes cluster level.

They have qos which can enforce some constraints https://github.com/Netflix/metaflow/blob/master/metaflow/plugins/kubernetes/kubernetes_decorator.py#L45