Parameter-Efficient Fine-Tuning with LoRA - Optimizing LLMs for Local GPU Training
Master PEFT LoRA techniques to efficiently fine-tune large language models on local GPUs through low-rank adaptation, reducing memory requirements while maintaining model performance.
Experimental KERAS Feature Space for Medical AI Heart Attack Prediction
Dive into advanced feature scaling with KERAS Feature Space, building a medical AI system to predict heart attack probabilities using structured data classification and optimized data pipelines.
Cost-Efficient Distributed Learning
Discover strategies for optimizing distributed gradient descent algorithms by managing worker responses and implementing adaptive techniques to balance computational efficiency with model quality in machine learning systems.
Adaptive Threading: The Goldilocks Dilemma with GC Parallelism
Dive into Eclipse OpenJ9's GC internals, exploring adaptive threading optimization and parallelism techniques to minimize garbage collection pause times and enhance Java application performance.
Infrastructure and Tooling - Full Stack Deep Learning - March 2019
Explore infrastructure and tooling for deep learning projects, covering GPU machines, experiment tracking, and all-in-one solutions like AWS SageMaker.
Efficient AutoML with Ludwig, Ray, and Nodeless Kubernetes
Explore efficient AutoML using Ludwig and Ray, leveraging heuristics for model production. Learn how Nodeless Kubernetes optimizes GPU resource management, reducing costs in cloud environments.
Resource Orchestration of HPC on Kubernetes - Where We Are Now and the Journey Ahead
Explore NUMA-aware scheduling in Kubernetes for HPC workloads, addressing challenges, solutions, and future developments in resource orchestration and performance optimization.
Kueue: Kubernetes-Native Job Queueing for Batch Workloads
Explore Kueue, a Kubernetes-native job queueing solution for efficient resource management and fair sharing in batch workloads, addressing limitations in pod-centric scheduling.
Volcano: Introduction and Deep Dive into High-Performance Workload Management on Kubernetes
Explore Volcano: a powerful batch scheduling system for high-performance workloads on Kubernetes. Learn key concepts, features, and real-world use cases for ML/DL, big data, and bioinformatics applications.
KubeFlux - An HPC Scheduler Plugin for Kubernetes
Explore KubeFlux, a Kubernetes plugin integrating HPC scheduling capabilities. Learn about its performance benefits and potential for enhancing cloud-based high-performance computing workloads.
KubeVirt v0.53 - New Features for Durability, Security, and Pluggability
Explore KubeVirt v0.53's enhancements in scalability, security, storage, and networking. Learn about new features for high-performance computing, unprivileged VMs, and improved automation for cloud-native virtualization.
Five Ways to Understand Kubernetes Networking for Performance-Intensive Workloads
Explore Kubernetes networking innovations for high-performance workloads. Compare CNI implementations using real-world benchmarks to analyze performance in scientific computing and machine learning.
Innovations in Theoretical Computer Science 2020 - Session 5
Explore cutting-edge research in theoretical computer science, including secure computation, communication protocols, coding theory, and fault-tolerant algorithms presented at ITCS 2020.
Modular Language Models: Addressing Limitations of Dense Training
Explore modular language models: specialized components for distinct domains, enabling customization, parallelization, and sparse activation. Discover a new approach to collaborative LM development.
How openEuler is Powering Tomorrow's AI and Smart Infrastructure
Explore openEuler's innovations in AI infrastructure and heterogeneous computing, including cloud-native solutions and openEuler NEXT, designed to optimize AI workflows and enhance user experience.
Scheduling Array Operations for GPU and Distributed Computing
Explore array operation scheduling for optimized execution across hardware. Learn about static scheduling, type theory, and parallelization for GPUs and distributed computing.
Report on Co-dfns: Current State, Developments, and Future Outlook - Dyalog '22
Explore the latest developments in Co-dfns, a high-performance compiler for parallel dfns, and its impact on the computing landscape. Learn about its benefits and future prospects.
Future(s) - Asynchronous Programming in Dyalog - Part 3a
Explore asynchronous programming in Dyalog, including green threads, .NET tasks, and futures. Learn about unifying I-beam functions and see demonstrations using N-queens and wait functions.
Multi-Stage Vertex-Centric Programming for Agent-Based Simulations
Explore multi-stage vertex-centric programming for agent-based simulations, improving performance and control flow in complex graph algorithms through meta-programming techniques.
MultiGPU Performance-Portable Solution for Array Programming Based on Kokkos
Explore a new Kokkos-based API for transparent, portable multiGPU array computation. Learn about implementation variations, autotuning, and performance comparisons with MPI+Kokkos on top supercomputers.
In-Order Execution Nails Every Weak Memory Behavior
Explore in-order execution's comprehensive coverage of weak memory behaviors, unveiling its significance in modern computing systems.