aws-samples/sample-rlinf-on-eks

gpu-instance-selection

Select optimal EC2 GPU instance types for Physical AI RL training based on VRAM, interconnect, and cost requirements

Langue du texte source : anglais

kubernetes-native-features

Leverage Kubernetes-native features for GPU training including GPU Operator, topology scheduling, and priority classes

Langue du texte source : anglais

deployment-method

Deploy and manage Physical AI RL training jobs on EKS including CodeBuild CI, GitOps, and experiment management

Langue du texte source : anglais

eks-cluster-provisioning

Provision an Amazon EKS cluster with GPU node groups, EFA networking, and OIDC for Physical AI RL training workloads

Langue du texte source : anglais

rlinf-dataset-preparation

RLinf: Download, preprocess, and stage datasets and model weights for RLinf training on EKS

Langue du texte source : anglais

functional-testing

Validate the Physical AI RL training stack end-to-end from infrastructure smoke tests to training step verification

Langue du texte source : anglais

kubernetes-manifests

Construct Kubernetes Jobs, StatefulSets, and Services for GPU-accelerated RL training workloads on EKS

Langue du texte source : anglais

rlinf-functional-testing

RLinf: Validate RLinf training examples end-to-end from container tests to single-step training

Langue du texte source : anglais

rlinf-open-source-plugins

RLinf: Integrate Ray, veRL, vLLM, and KubeRay into the RLinf training stack on EKS

Langue du texte source : anglais

ami-and-node-configuration

Configure AMI selection and GPU node bootstrap including NVIDIA drivers, EFA, and Lustre client

Langue du texte source : anglais

container-image-building

Build container images for Physical AI RL training including dependency conflict management, patching, and CodeBuild CI

Langue du texte source : anglais

monitoring-observability

Monitor GPU utilization, training progress, and costs using DCGM, Prometheus, Grafana, MLflow, and CloudWatch

Langue du texte source : anglais

networking-efa

Configure Elastic Fabric Adapter networking for high-performance NCCL allreduce in multi-node GPU training

Langue du texte source : anglais

rlinf-benchmarking

RLinf: Evaluate training quality and infrastructure performance using LIBERO, RoboTwin, and throughput benchmarks

Langue du texte source : anglais

rlinf-pytorch-training-scripts

RLinf: Adapt RLinf training scripts for distributed execution on Amazon EKS with veRL, Ray, and FSDP

Langue du texte source : anglais

storage-fsx-lustre

Deploy Amazon FSx for Lustre as shared high-performance storage for datasets, checkpoints, and model weights

Langue du texte source : anglais

storage-mountpoint-s3

Mount S3 buckets as POSIX-like filesystems in training pods via the Mountpoint for Amazon S3 CSI driver

Langue du texte source : anglais

storage-s3-connector-pytorch

Use Amazon S3 Connector for PyTorch to read datasets and save checkpoints directly to S3

Langue du texte source : anglais

storage-s3-files

Mount S3 buckets as native NFS file systems on EKS via Amazon S3 Files for shared, low-latency file access

Langue du texte source : anglais