📄️ Introduction
Running AI/ML platforms on Kubernetes can greatly simplify and automate the deployment, scaling, and management of these complex applications. There are a number of popular tools and technologies that have emerged to support this use case, including TensorFlow, PyTorch, Ray, MLFlow, etc.
📄️ Ray on EKS
This blueprint should be considered as experimental and should only be used for proof of concept.
📄️ JupyterHub on EKS
JupyterHub s a powerful multi-user server that enables users to access and interact with Jupyter notebooks and other Jupyter-compatible environments. It provides a collaborative platform where multiple users can simultaneously access and utilize notebooks, fostering collaboration and sharing among users. JupyterHub allows users to create their own isolated computing environments (known as "spawners") and launch Jupyter notebooks or other interactive computing environments within those environments. This provides each user with their own workspace, including files, code, and computational resources.
📄️ EMR NVIDIA Spark-RAPIDS
The NVIDIA RAPIDS Accelerator for Apache Spark is a powerful tool that builds on the capabilities of NVIDIA CUDA® - a transformative parallel computing platform designed for enhancing computational processes on NVIDIA's GPU architecture. RAPIDS, a project developed by NVIDIA, comprises a suite of open-source libraries that are hinged upon CUDA, thereby enabling GPU-accelerated data science workflows.
📄️ Trainium on EKS
AWS Trainium is an advanced ML accelerator that transforms high-performance deep learning(DL) training. Trn1 instances, powered by AWS Trainium chips, are purpose-built for high-performance DL training of 100B+ parameter models. Meticulously designed for exceptional performance, Trn1 instances cater specifically to training popular Natual Language Processing(NLP) models on AWS, offering up to 50% cost savings compared to GPU-based EC2 instances. This cost efficiency makes them an attractive option for data scientists and ML practitioners seeking optimized training costs without compromising performance.