The Open Hackathons program presents a unique opportunity for scientists, researchers, and developers to build collaborations and learn the hands-on skills needed to accelerate and optimize their applications using a variety of programming models, libraries and tools. Currently, we offer two types of events: Hackathons and Bootcamps.
Bootcamps are designed to help build confidence and eventually prepare developers to participate in Hackathons This repository contains bootcamp material for high performance computing (HPC), Artificial Intelligence (AI), and convergence of both.
- HPC :: This bootcamp content focuses on how to follow the “Analyze, Parallelize, and Optimize” cycle to write parallel codes using different parallel programming models for accelerating HPC simulations.
| Lab | Description |
|---|---|
| N-Ways | This Bootcamp will cover multiple GPU programming models and choose the one that best fits your needs. The material supports different programming langauges including C ( CUDA C, OpenACC C, OpenMP C, C++ stdpar ), Fortran ( CUDA Fortran, OpenACC Fortran, OpenMP Fortran, ISO DO CONCURRENT ) Python ( Numba, CuPy ) |
| Multi GPU Programming Model | This bootcamp will cover scaling applications to multiple GPUs across multiple nodes. Moreover, understanding of the underlying technologies and communication topology will help us utilize high-performance NVIDIA libraries to extract more performance out of the system |
| HPC Profiler | This repository contains learning materials and exercises for NVIDIA Nsight Tools. Goal is to learn how to profile your application with NVIDIA Nsight Systems,Compute and NVTX API calls to find performance limiters and bottlenecks and apply incremental parallelization strategies |
| C++ Standard Parallelism | This repository contains learning materials and exercises for C++ Standard Parallelism. The content covers step by steps of accelerating a portable HPC application with CPUs and GPUs using the parallelism and concurrency features of the C++17 and C++20 standards. Goal is to learn about C++ concurrency features like threads, atomics, barriers, and parallel algorithms. This forked content has been tested. The owners of the material are Gonzalo Brito and Jonas Latt |
- Convergence of HPC and AI :: The bootcamp content focuses on how AI can accelerate HPC simulations by introducing concepts of Deep Neural Networks, including Physics Informed Neural Networks, neural operators and techniques to visualise physically acccurate outputs from the trained models.
| Lab | Description |
|---|---|
| End to End AI for Science | This bootcamp provides a step-by-step overview of the fundamentals of deep neural networks, walks attendees through the hands-on experience of building and improving deep learning models using a framework that uses the fundamental laws of physics to model the behavior of complex systems, and enables attendees to visualize the physically accurate outputs of the trained model in near real-time. |
- AI :: The bootcamp content focuses on using popular accelerated AI frameworks and using optimization techniques to get max performance from accelerators like GPU.
| Lab | Description |
|---|---|
| End to End Computer Vision | This repository contains the material for the end-to-end computer vision bootcamp, the goal of which is to build a complete end-to-end computer vision pipeline for an object detection application. This bootcamp will introduce participants to multiple NVIDIA® SDKs, most notably NVIDIA TAO Toolkit, NVIDIA TensorRT™, NVIDIA Triton™ Inference Server, and NVIDIA DeepStream SDK. Participants will also have hands-on experience in data preprocessing, model training, optimization, and deployment at scale |
| AI Profiler | Learn how optimizing a Deep Neural Network (DNN) training using profiling tools |
| End to End LLM | The End-to-End LLM (Large Language Model) Bootcamp is designed from a real-world perspective that follows the data processing, development, and deployment pipeline paradigm. Attendees walk through the workflow of preprocessing the openassistant-guanaco dataset for the Text Generation task and training the dataset using the LLAMA 2 7Billion Model,a pre-trained and fine-tuned LLM. Attendees will also learn to optimize an LLM using NVIDIA® TensorRT™ LLM, an SDK for high-performance large language model inference, understand guardrail prompts and responses from the LLM model using NVIDIA NeMo Guardrails, and deploy the AI pipeline using NVIDIA TensorRT LLM Backend (powered by Triton™ Inference Server), an open-source software that standardizes LLM deployment and execution across every workload. |
Each lab contains docker and singularity definition files. Follow the readme files inside each on how to build the container and run the labs inside it.
- The repository uses Apache 2.0 license. For more details on folder structure developers may refer to CONTRIBUTING.md file.
- Bootcamp users may request for newer training material or file a bug by filing a github issues
- Please do go through the existing list of issues to get more details of upcoming features and bugs currently being fixed)
- All materials developed are tested with latest GPU Architectures (V100, A100). Most labs unless specified explicitly are expected to work even on older GPU architectures and with lesser compute and memory capacity like the one present even in laptops. There will be change in performance results observed based on GPU used. In case you see any issue using the material on other GPU please file an issue in Github mentioning the details of GPU and CUDA Driver version installed.
- The material developed are tested inside container environment like Docker and Singularity. In case the users don't have container environment in the cluster, they can explicitly look at the steps mentioned in the Dockerfile and Singularity scripts and install the dependenciesmanually.
- Most bootcamps are jupyter based and by default the Dockerfile and Singularity script runs the jupyter notebook at port 8888. In a munti-tenancy environment the admins are requested to explicitly map the ports to individual users else will result into port conflict issues. We recommend having installations of interactive interface to remote computing resources like Open OnDemand or JupyterHub coupled with scheduler (SLURM, Kubernetes etc ) to do these resources mapping automatically.
Please join OpenACC Slack Channel.