I will share the technical hurdles and lessons learned from implementing Machine Learning training with cloud-based pipelines using GitLab CICD, Docker+Machine, and GPU-powered virtual machines in the cloud.
Data Science infrastructure requirements are easily capable of racking up high costs with GPU based virtual-machines running 24/7 in the cloud, or require specialised hardware to be maintained and powered on-prem. Without huge expense, achieving agile ML model development can be hard.
Data Scientists in the organization can now quickly build their own pipelines with high agility and speed, at reduced infrastructure costs than before, and time reduced in debugging issues.