Build consistent and portable ML environments with containers
summary
Prototype fine‑tuned a DistilBERT text classifier (star ratings) using Hugging Face tokenizers and TFRecords, then packages the training code in Docker, runs it on Amazon EKS (with FSx for Lustre mounted) and on Amazon SageMaker via the TensorFlow estimator. Training freezes the BERT base and fine‑tunes the classifier head, yields ~25% test accuracy in notebook, EKS and SageMaker, and demonstrates prebuilt AWS Deep Learning Containers and 61% cost savings with SageMaker Spot
AI-written summary · about
chapters (9)
- 0:00Start
- 0:50Use Case
- 5:10How To Use Containers with Amazon Eks
- 7:07Training Yml
- 9:15Script Mode
- 12:14Prepare the Data
- 15:25Model Training Code
- 23:37Sagemaker Python Sdk
- 28:53Summary
related talks
45mBuild NLP Models with Amazon SageMaker
35mAccelerate deep learning in the cloud with custom ML environments
39mTrain ML models at scale with Amazon SageMaker Training
45mAutomate your ML workflows with pipelines
19mDeveloping large-scale machine learning algorithms on Amazon SageMaker
54mRipped Machine Learning with Distributed Computing
24mDocker for Machine Learning!
19mUsing Amazon SageMaker to develop and deploy ML models