Automate your ML workflows with pipelines
summary
Builds an end-to-end ML pipeline using Amazon SageMaker Pipelines that preprocesses text, generates embeddings, fine-tunes a pre-trained BERT model with a TensorFlow estimator, captures validation metrics and applies step caching. The demo uses a 90/5/5 split and a single training epoch for speed, registers models in a model registry with manual approval, triggers CI/CD via CodeCommit/CodeBuild/CodePipeline, deploys to a staging endpoint and returns live predictions successfully.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 1:59What is MLOps
- 5:54Building resilient ML pipelines
- 8:01Manual ML workflow challenges
- 9:03Demo: Model training pipeline
- 20:44Evaluating model performance
- 22:28Quality gates for model approval
- 26:29End-to-end pipeline steps
- 30:23Automating quality gates
- 31:47CI/CD pipeline triggers
- 32:07Demo: SageMaker Projects
- 43:22Building resilient AI solutions
related talks
29mHow to create fully automated ML workflows with Amazon SageMaker Pipelines
34mImplement MLOps practices with Amazon SageMaker
34mImplement MLOps practices with Amazon SageMaker
29mImplementing MLOps practices with Amazon SageMaker
18mAutomate the boring ML stuff with pipelines
43mMachine Learning Ops: DevOps for Data Science
33mBuild, train, and deploy ML models using Amazon SageMaker
29mAutomatically Build Models With Amazon Sagemaker