Table of Contents

---

Session 1 Hands-on Labs

• data_parallel_pytorch_code_changes points basic code changes when you bring a PyTorch script to use SageMaker distributed training library data parallelism.

• data_parallel_tensorflow_code_changes points basic code changes when you bring a Tensorflow script to use SageMaker distributed training library data parallelism.

• huggingface_data_parallelism_checkpointing shows how to leverage SageMaker checkpointing to save transformer checkpoints during SageMaker distribued training (data parallelism) and resume training from checkpoints.

• huggingface_data_parallelism_incremental_training shows how to do SageMaker distributed training (data parallelism) with HuggingFace framework and how to do incremental training from a saved model artifact.

---

Session 2 Hands-on Labs

• byos_pytorch takes PyTorch framework as an example to show how to bring your own script to train and deploy a model on SageMaker.

• byoc_pytorch shows how to extend AWS pre-built deep learning container (PyTorch as an example) to build your own container and bring it to SageMaker for model training.

• xgboost_builtin_distributed shows doing distributed training with SageMaker built-in XgBoost algorithm, and using SageMaker automatic model tuning to tune model hyperparameters.

• xgboost_script_mode_distributed shows how to leverage pre-built XgBoost framework container to train a XgBoost model in a distributed training fashion.

• xgboost_pyspark shows using SageMaker pre-built Spark container to train a XgBoost model. Note: notebook is tested on SageMaker classic notebook instance.

Other sample codes

• feature_store Example use case with Offline Feature Store SDK and create dataset

• spark_distributed_data_processing Example use case with Distributed Data Processing using Apache Spark and SageMaker Processing

• sagemaker_pipelines Example use case with SageMaker Pipelines which includes Processing, Training, Evaluation, Condition and Model Registry Steps