This repository contains examples and related resources for Amazon SageMaker Training jobs over different instance types focusing on the aspects of time to train and cost to train.
Amazon SageMaker makes it easy to train machine learning using EC2 instances. There are many instance types to choose from and this choice affects the speed and cost of training. This repository contains example benchmark for various deep learning use cases. You can see results directly in the notebook, reproduce results by re-running the notebooks. And alter the notebooks to create new scenarios to benchmark.
The repository contains the following resources:
-
G5 instance types benchmarks:
- ml.g5.2xlarge vs. ml.p3.2xlarge for training RoBERTa NLP model - This notebook shows how to evaluate which instance to use to train faster and chaper. It includes measures time-to-train and
Raise an issue on this repo.
See CONTRIBUTING for more information.
This library is licensed under the Apache License 2.0 License. See the LICENSE file.