1

Mercury: Unlocking Multi-GPU Operator Optimization for LLMs via Remote Memory Scheduling
Remote memory scheduling framework that optimizes LLM operators across multi-GPU deployments.
Cite DOI PDF
Comprehensive Evaluation of LLMs in HPC Code Performance Optimization
Benchmarks and evaluates large language models for optimizing high-performance computing code.
Cite arXiv
KPerfIR: Towards an Open and Compiler-centric Ecosystem for GPU Kernel Performance Tooling on Modern AI Workloads
An open, compiler-focused infrastructure for profiling and optimizing GPU kernels on AI workloads.
Cite DOI PDF arXiv
DeepContext: A Context-aware, Cross-platform, and Cross-framework Tool for Performance Profiling and Analysis of Deep Learning Workloads
Performance profiling toolkit that unifies deep learning workload analysis across platforms and frameworks.
Cite DOI arXiv
Triton-Viz: Visualizing GPU Programming in AI Courses
GPU programming is a critical component in AI system courses, which is notoriously difficult to learn and teach, given its unique …
Cite Project URL
SS1: Accelerating Inference with Fast and Expressive Sketch Structured Transform
Tensor multiplication with learned weight matrices is the fundamental building block in deep learning models. These matrices can often …
Cite Project
Centimani: Enabling Fast AI Accelerator Selection for DNN Training with a Novel Performance Predictor
For an extended period, graphics processing units (GPUs) have stood as the exclusive choice for training deep neural network (DNN) …
Cite Project URL
FASTEN: Fast GPU-accelerated Segmented Matrix Multiplication for Heterogenous Graph Neural Networks
This paper introduces FASTEN, a cutting-edge library developed to address the computational challenges inherent in Heterogeneous Graph …
Cite Project DOI URL
PyTorch 2: Faster Machine Learning Through Dynamic Python Bytecode Transformation and Graph Compilation
This paper introduces two extensions to the popular PyTorch machine learning framework, TorchDynamo and TorchInductor, which implement …
Cite Project DOI URL
Hardware-Aware Compression with Random Operation Access Specific Tile (ROAST) Hashing
Advancements in deep learning are often associated with increasing model sizes. Training and deploying large models require …
Cite Project URL