A GPU kernel library and generator for LLM inference, providing unified APIs for attention, GEMM, MoE, and sampling operations.
awslabs/awsome-distributed-ai
Reference architectures, templates and examples for running distributed AI training and inference on AWS clusters.
- Stars
- 495
- Forks
- 215
- Open issues
- 38
- Last commit
- 23 hours ago
What is it?
This repository collects reference architectures, deployment templates and runnable examples for distributed AI training and inference on AWS. It covers cluster platforms including Amazon SageMaker HyperPod, AWS ParallelCluster, AWS Parallel Computing Service and Amazon EKS, with CloudFormation and Terraform templates for each.
The examples span training frameworks such as PyTorch DDP and FSDP, Megatron-LM and NeMo, and serving engines such as vLLM, SGLang and NVIDIA Dynamo. The repo also includes Packer and Ansible scripts for building custom machine images, cluster health validation tools, EFA and Nsight observability guides, and micro-benchmarks for NCCL, NCCOM and NVSHMEM.
Teams running multi-node GPU training or inference on AWS can use it as a starting point for cluster setup, launch scripts and performance tuning.
Key features
- 01Reference architectures for SageMaker HyperPod, ParallelCluster, PCS and EKS
- 02Training examples for PyTorch FSDP, Megatron-LM and NeMo
- 03Inference examples for vLLM, SGLang and NVIDIA Dynamo
- 04Packer and Ansible scripts for custom AMIs
- 05GPU cluster health checks and EFA monitoring exporters
- 06Micro-benchmarks for NCCL, NCCOM and NVSHMEM
Use cases
- Setting up a Slurm or Kubernetes GPU cluster on AWS for large model training
- Deploying multi-node LLM inference with vLLM or SGLang on EKS
- Validating GPU and network performance before running long training jobs
- Profiling distributed workloads with Nsight and EFA metrics
Quick start
- 1
Browse the examples directory to choose a reference architecture or example
- 2
Read the README of the chosen architecture or example for setup instructions
- 3
Clone the repository to your local machine
git clone https://github.com/awslabs/awsome-distributed-ai.git
Things to know
- Targets AWS-specific clusters (SageMaker HyperPod, ParallelCluster, PCS, EKS), so it is not useful without an AWS account and cluster setup.
- It is a collection of reference architectures and examples rather than a single installable package, so there is no one-step install.
In collections
Topics
FAQ
- What is awsome-distributed-ai?
- Reference architectures, templates and examples for running distributed AI training and inference on AWS clusters. This repository collects reference architectures, deployment templates and runnable examples for distributed AI training and inference on AWS. It covers cluster platforms including Amazon SageMaker HyperPod, AWS ParallelCluster, AWS Parallel Computing Service and Amazon EKS, with CloudFormation and Terraform templates for each.
- Is awsome-distributed-ai free to use? What license does it have?
- awsome-distributed-ai is open source under the MIT-0 license. Check the license terms before using it in commercial products.
- Is awsome-distributed-ai still maintained?
- Actively maintained. The last commit to awsome-distributed-ai was on Nov 6, 2026.
- What can I use awsome-distributed-ai for?
- Setting up a Slurm or Kubernetes GPU cluster on AWS for large model training Deploying multi-node LLM inference with vLLM or SGLang on EKS Validating GPU and network performance before running long training jobs Profiling distributed workloads with Nsight and EFA metrics
- How do I install awsome-distributed-ai?
- Browse the examples directory to choose a reference architecture or example Read the README of the chosen architecture or example for setup instructions Clone the repository to your local machine: git clone https://github.com/awslabs/awsome-distributed-ai.git
- Where can I get awsome-distributed-ai?
- The source code and installation instructions are on GitHub: https://github.com/awslabs/awsome-distributed-ai
Summary generated by AI from the project's README. GitHub stats refresh daily.
Related projects
Kubernetes-native platform for distributed AI model training and LLM fine-tuning across PyTorch, JAX, XGBoost and other frameworks.
NCCL is NVIDIA's library of optimized GPU communication primitives such as all-reduce and all-gather for single- and multi-node training.
An open book of methods, scripts and step-by-step guides for training, fine-tuning and running inference on large language and multimodal models.
An LLM inference engine for Hugging Face-compatible models, built on vLLM, with OpenAI-compatible serving and distributed deployment.
A GPU communication library for expert-parallel MoE dispatch and combine, with FP8 dispatch and NVLink/RDMA support for training and inference.