Understanding the Performance Horizon of the Latest ML Workloads with NonGEMM Workloads

• URL: http://arxiv.org/abs/2404.11788v5
• Date: Wed, 16 Apr 2025 05:47:53 GMT
• Title: Understanding the Performance Horizon of the Latest ML Workloads with NonGEMM Workloads
• Abstract summary: We conduct a performance analysis of 17 widely adopted ML models in Hugging Face and Torchvision.<n>Non-GEMM performance bottleneck accounts for 11.3% to 73.6% of total latency, on average.<n>We demystify the most dominant non-GEMM operators for each model and deployment software.
• Score: 1.4107658666233678
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Among ML operators today, GEneralMatrix Multiplication (GEMM)-based operators are known to be key operators that build the main backbone of ML models. As their computational overhead dominates the overall execution time (e.g., 42.8% - 96.6% in our results), GEMM operators have been the prime optimization targets for fast ML inference. This led to advanced GPUs and accelerators available today, which provided significant boost in the GEMM performance compared to CPUs, aligned with the lesson from Amdahl's law. However, accelerating GEMM has significantly shifted the Amdahl's law's landscape for ML inference; due to the decreased GEMM execution time, the relative execution time of non-GEMM operators is now significant. Although the importance of non-GEMM performance is increasing, we have little knowledge about the non-GEMM performance horizon in the latest hardware platforms and models. Therefore, to guide non-GEMM-oriented optimizations, we conduct a thorough performance analysis of 17 widely adopted ML models in Hugging Face and Torchvision on workstation and data center platforms with/without GPUs. We discover that non-GEMM performance bottleneck is a considerable issue across all the platforms and models, accounting for 11.3% to 73.6% of total latency, on average. The challenge significantly aggravates when we apply quantization, which is a common model compression technique, due to the boosted GEMM performance and extra non-GEMM operators for dequantization and requantization. To provide insights into non-GEMM optimization targets, we demystify the most dominant non-GEMM operators for each model and deployment software. We also show that widely adopted optimizations such as operator fusion do not completely address the non-GEMM performance bottleneck, where non-GEMM operators still account for 15% to 48% of total latency.

Related papers

• MMOE: Modernizing Diffusion Transformers with Efficient Expert Design [76.86793131833551]
  Modern large language models scale successfully by pairing capacity growth with efficiency, keeping per-token and deployment costs under control as capacity grows.<n>Recent efforts mostly enlarge total parameter counts and sparsity ratios without importing the efficiency mechanisms that made LLM scaling practical.<n>We introduce ModernMOE, a modernization of SiT-style diffusion transformers that adapts routed experts, shared and lightweight experts, gate-residual routing, and attention-residual information reuse to AIGC generation.
  arXiv  Detail & Related papers  (2026-07-27T17:05:04Z)
• CODA: Rewriting Transformer Blocks as GEMM-Epilogue Programs [48.961770224590204]
  CODA is a GPU kernel abstraction that expresses these computations as GEMM-plus-epilogue programs.<n>We show that CODA offers a practical path toward combining framework-level productivity with hardware-level efficiency.
  arXiv  Detail & Related papers  (2026-05-19T02:30:43Z)
• LP-GEMM: Integrating Layout Propagation into GEMM Operations [0.20388938295521575]
  This paper introduces LP-GEMM, a decomposition of the GEMM kernel that enables packing- propagation across sequential GEMM operations.<n>We show average speedups of 2.25x over OpenBLA on Intel x86 for sequential GEMMs and competitive gains relative to vendor-optimized libraries.
  arXiv  Detail & Related papers  (2026-04-06T11:24:34Z)
• Exploration of Unary Arithmetic-Based Matrix Multiply Units for Low Precision DL Accelerators [3.1197830253827026]
  General matrix multiplication (GEMM) is a fundamental operation in deep learning (DL)<n>Recent works have proposed novel unary GEMM designs as an alternative to conventional binary GEMM hardware.<n>We demonstrate how unary GEMM can be effectively used for energy-efficient compute in future edge AI accelerators.
  arXiv  Detail & Related papers  (2026-01-31T18:10:55Z)
• Nemotron-Flash: Towards Latency-Optimal Hybrid Small Language Models [97.55009021098554]
  This work aims to identify the key determinants of SLMs' real-device latency and offer generalizable principles and methodologies for SLM design and training.<n>We introduce a new family of hybrid SLMs, called Nemotron-Flash, which significantly advances the accuracy-efficiency frontier of state-of-the-art SLMs.
  arXiv  Detail & Related papers  (2025-11-24T08:46:36Z)
• Diagonally-Weighted Generalized Method of Moments Estimation for Gaussian Mixture Modeling [7.266474956854856]
  The diagonally-weighted GMM (DGMM) achieves a balance among statistical efficiency, computational complexity, and numerical stability.<n>We apply DGMM to study the parameter estimation problem for weakly separated heteroscedastic low-rank Gaussian mixtures.<n>In numerical studies, DGMM attains smaller estimation errors while requiring substantially shorter runtimes than MM and GMM.
  arXiv  Detail & Related papers  (2025-07-28T01:24:55Z)
• Modality-Balancing Preference Optimization of Large Multimodal Models by Adversarial Negative Mining [66.54211199959298]
  We propose a novel preference learning framework, Modality-Balancing Preference Optimization (MBPO), to address the modality imbalance in LMMs.<n>MBPO constructs a more effective offline preference dataset by generating hard negatives, i.e., rejected responses misled by LLM biases.<n>It can enhance LMM performance on challenging vision-language tasks and effectively reduce hallucinations.
  arXiv  Detail & Related papers  (2025-05-20T03:59:05Z)
• Reinforced Model Merging [53.84354455400038]
  We present an innovative framework termed Reinforced Model Merging (RMM), which encompasses an environment and agent tailored for merging tasks. By utilizing data subsets during the evaluation process, we addressed the bottleneck in the reward feedback phase, thereby accelerating RMM by up to 100 times.
  arXiv  Detail & Related papers  (2025-03-27T08:52:41Z)
• Frequency-Aware Masked Autoencoders for Human Activity Recognition using Accelerometers [0.1499944454332829]
  Supervised machine learning and deep learning algorithms have long been used to extract meaningful activity information from raw accelerometry data. We propose a novel spectrogram-based loss function named the log-scale mean magnitude (LMM) loss for human activity recognition. Our findings demonstrate that the LMM loss is a robust and effective method for pretraining MAE models on accelerometer data for HAR.
  arXiv  Detail & Related papers  (2025-02-17T14:57:51Z)
• Tackling the Dynamicity in a Production LLM Serving System with SOTA Optimizations via Hybrid Prefill/Decode/Verify Scheduling on Efficient Meta-kernels [12.77187564450236]
  We introduce XY-Serve, a versatile, Ascend native, end-to-end production large language model (LLM) serving system. The core idea is an abstraction mechanism that smooths out the workload variability by decomposing computations into fine-grained meta primitives. For GEMM, we introduce a virtual padding scheme that adapts to dynamic shape changes while using highly efficient GEMM primitives with assorted fixed tile sizes.
  arXiv  Detail & Related papers  (2024-12-24T02:27:44Z)
• Panacea: Novel DNN Accelerator using Accuracy-Preserving Asymmetric Quantization and Energy-Saving Bit-Slice Sparsity [2.78181759570722]
  Low bit-precisions and their bit-slice sparsity have been studied to accelerate general matrix-multiplications (GEMM) during large-scale deep neural network (DNN) inferences. Recent studies have actively utilized asymmetric quantization for activations without requiring additional operations. This paper proposes an Asymmetrically-Quantized bit-Slice GEMM for the first time.
  arXiv  Detail & Related papers  (2024-12-13T11:44:09Z)
• EPS-MoE: Expert Pipeline Scheduler for Cost-Efficient MoE Inference [49.94169109038806]
  This paper introduces EPS-MoE, a novel expert pipeline scheduler for MoE that surpasses the existing parallelism schemes. Our results demonstrate at most 52.4% improvement in prefill throughput compared to existing parallel inference methods.
  arXiv  Detail & Related papers  (2024-10-16T05:17:49Z)
• VisualAgentBench: Towards Large Multimodal Models as Visual Foundation Agents [50.12414817737912]
  Large Multimodal Models (LMMs) have ushered in a new era in artificial intelligence, merging capabilities in both language and vision to form highly capable Visual Foundation Agents. Existing benchmarks fail to sufficiently challenge or showcase the full potential of LMMs in complex, real-world environments. VisualAgentBench (VAB) is a pioneering benchmark specifically designed to train and evaluate LMMs as visual foundation agents.
  arXiv  Detail & Related papers  (2024-08-12T17:44:17Z)
• Revisiting Zeroth-Order Optimization for Memory-Efficient LLM Fine-Tuning: A Benchmark [166.40879020706151]
  This paper proposes a shift towards BP-free, zeroth-order (ZO) optimization as a solution for reducing memory costs during fine-tuning. Unlike traditional ZO-SGD methods, our work expands the exploration to a wider array of ZO optimization techniques. Our study unveils previously overlooked optimization principles, highlighting the importance of task alignment, the role of the forward gradient method, and the balance between algorithm complexity and fine-tuning performance.
  arXiv  Detail & Related papers  (2024-02-18T14:08:48Z)
• Large Language Model for Multi-objective Evolutionary Optimization [26.44390674048544]
  Multiobjective evolutionary algorithms (MOEAs) are major methods for solving multiobjective optimization problems (MOPs) Recent attempts have been made to replace the manually designed operators in MOEAs with learning-based operators. This work investigates a novel approach that leverages the powerful large language model (LLM) to design MOEA operators.
  arXiv  Detail & Related papers  (2023-10-19T07:46:54Z)
• Merging Experts into One: Improving Computational Efficiency of Mixture of Experts [71.44422347502409]
  A sparse Mixture of Experts (MoE) can reduce the cost by activating a small subset of parameters. Can we retain the advantages of adding more experts without substantially increasing the computational costs? We propose a computation-efficient approach called textbftexttMerging Experts into One (MEO) which reduces the computation cost to that of a single expert.
  arXiv  Detail & Related papers  (2023-10-15T13:28:42Z)
• Use Your INSTINCT: INSTruction optimization for LLMs usIng Neural bandits Coupled with Transformers [66.823588073584]
  Large language models (LLMs) have shown remarkable instruction-following capabilities and achieved impressive performances in various applications. Recent work has used the query-efficient Bayesian optimization (BO) algorithm to automatically optimize the instructions given to black-box LLMs. We propose a neural bandit algorithm which replaces the GP in BO by an NN surrogate to optimize instructions for black-box LLMs.
  arXiv  Detail & Related papers  (2023-10-02T02:01:16Z)
• Relational Reasoning via Set Transformers: Provable Efficiency and Applications to MARL [154.13105285663656]
  A cooperative Multi-A gent R einforcement Learning (MARL) with permutation invariant agents framework has achieved tremendous empirical successes in real-world applications. Unfortunately, the theoretical understanding of this MARL problem is lacking due to the curse of many agents and the limited exploration of the relational reasoning in existing works. We prove that the suboptimality gaps of the model-free and model-based algorithms are independent of and logarithmic in the number of agents respectively, which mitigates the curse of many agents.
  arXiv  Detail & Related papers  (2022-09-20T16:42:59Z)
• MLGOPerf: An ML Guided Inliner to Optimize Performance [7.314201117946244]
  This paper presents the first end-to-end framework capable of optimizing performance using LLVM's ML-Inliner. It employs a secondary ML model to generate rewards used for training a retargeted Reinforcement learning agent. It does so by predicting the post-inlining speedup of a function under analysis and it enables a fast training framework for the primary model.
  arXiv  Detail & Related papers  (2022-07-18T05:47:29Z)
• Image Modeling with Deep Convolutional Gaussian Mixture Models [79.0660895390689]
  We present a new formulation of deep hierarchical Gaussian Mixture Models (GMMs) that is suitable for describing and generating images. DCGMMs avoid this by a stacked architecture of multiple GMM layers, linked by convolution and pooling operations. For generating sharp images with DCGMMs, we introduce a new gradient-based technique for sampling through non-invertible operations like convolution and pooling. Based on the MNIST and FashionMNIST datasets, we validate the DCGMMs model by demonstrating its superiority over flat GMMs for clustering, sampling and outlier detection.
  arXiv  Detail & Related papers  (2021-04-19T12:08:53Z)
• A Framework of Inertial Alternating Direction Method of Multipliers for Non-Convex Non-Smooth Optimization [17.553531291690025]
  We propose an algorithmic framework dubbed alternating methods of multipliers (iADMM) for solving a class of non nonsmooth multiblock composite problems. Our framework employs the general-major surrogateization (MM) principle to update each block of variables to unify the convergence analysis of previous ADMM schemes.
  arXiv  Detail & Related papers  (2021-02-10T13:55:28Z)
• Cauchy-Schwarz Regularized Autoencoder [68.80569889599434]
  Variational autoencoders (VAE) are a powerful and widely-used class of generative models. We introduce a new constrained objective based on the Cauchy-Schwarz divergence, which can be computed analytically for GMMs. Our objective improves upon variational auto-encoding models in density estimation, unsupervised clustering, semi-supervised learning, and face analysis.
  arXiv  Detail & Related papers  (2021-01-06T17:36:26Z)
• FusedMM: A Unified SDDMM-SpMM Kernel for Graph Embedding and Graph Neural Networks [3.577310844634503]
  We develop a fused matrix multiplication kernel that unifies sampled dense-dense matrix multiplication and sparse-dense matrix multiplication under a single operation called FusedMM. By using user-defined functions, FusedMM can capture almost all computational patterns needed by popular graph embedding and GNN approaches.
  arXiv  Detail & Related papers  (2020-11-07T18:06:57Z)
• The Performance Analysis of Generalized Margin Maximizer (GMM) on Separable Data [45.4329219943791]
  "Generalized Margin Maximizer", GMM, minimizes any arbitrary convex function of the parameter vector. We provide a precise analysis of the performance of GMM via the solution of a system of nonlinear equations. Our theoretical results are validated by extensive simulation results across a range of parameter values, problem instances, and model structures.
  arXiv  Detail & Related papers  (2020-10-29T06:40:05Z)
• Q-GADMM: Quantized Group ADMM for Communication Efficient Decentralized Machine Learning [66.18202188565922]
  We propose a communication-efficient decentralized machine learning (ML) algorithm, coined QGADMM (QGADMM) We develop a novel quantization method to adaptively adjust modelization levels and their probabilities, while proving the convergence of QGADMM for convex functions.
  arXiv  Detail & Related papers  (2019-10-23T10:47:06Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.