ML workloads on Kubernetes with Kubeflow: Why and How? - Alexander Krasilnikov

This talk addresses the challenges businesses face in adopting Kubernetes, particularly due to a lack of in-house skills necessary for managing Kubernetes-based infrastructures, especially in Machine Learning contexts. The speaker introduces essential Kubernetes constructs and their effective use for ML workloads, highlighting the open-source project Kubeflow, which enables users to harness Kubernetes for training and serving ML models. The session emphasizes Charmed Kubeflow, a Canonical distribution, demonstrating how it simplifies the deployment of large machine learning models. Additionally, attendees will learn about utilizing Kubeflow's finetuning pipeline for LLM finetuning in retrieval-augmented generation applications using Couchbase.