In this new video, I demo the newly launched Amazon SageMaker Training Compiler, a feature of SageMaker that can accelerate the training of deep learning (DL) models by up to 50% through more efficient use of GPU instances.
Starting from a couple of sample notebooks based on Hugging Face models (BERT and GPT-2), I train both vanilla and compiled jobs, and I compare their performance.
⭐️⭐️⭐️ Don't forget to subscribe to be notified of future videos ⭐️⭐️⭐️
Code: github.com/aws/amazon-sagemaker-examples/tree/master/sagemaker-training-compiler/huggingface
More Hugging Face on SageMaker notebooks: github.com/huggingface/notebooks/tree/master/sagemaker
New to Transformers? Check out the Hugging Face course at huggingface.co/course
- AWS re:Invent 2021 - A first look at SageMaker Training Compiler ( Download)
- Speeding up deep learning training with SageMaker Training Compiler ( Download)
- AWS re:Invent 2021: AWS On Air ft. Amazon SageMaker Studio Lab | AWS Events ( Download)
- AWS re:Invent 2021 - {New Launch} Introducing AWS Trainium-based Amazon EC2 Trn1 instances ( Download)
- AWS 2022 Machine Learning / New Algorithms - AWS re:invent 2021 ( Download)
- AWS re:Invent 2021 - Large-scale distributed training of media ML models with Amazon FSx ( Download)
- Introducing SageMaker Data Parallelism - AWS re:Invent 2020 ( Download)
- AWS re:Invent 2020: Machine learning inference with Amazon EC2 Inf1 instances ( Download)
- AWS re:Invent 2020: Accelerate data preparation with Amazon SageMaker Data Wrangler ( Download)
- AWS On Air ft. Amazon SageMaker Feature Store ( Download)
- Accelerate Transformers on Amazon SageMaker with AWS Trainium and AWS Inferentia ( Download)
- AWS reInvent 2021 Recap - Brooke Jamieson ( Download)
- AWS Partner Webinar: Object2Vec on Amazon SageMaker ( Download)
- How to Accelerate Your Models to Production with Amazon SageMaker ( Download)
- AWS re:Invent 2021 - Live from re:Invent: Tuesday Morning Update | AWS Events ( Download)
