Carole-Jean Wu

Mark Mazumder, Colby Banbury, Xiaozhe Yao et al.

Machine learning research has long focused on models rather than datasets, and prominent datasets are used for common ML tasks without regard to the breadth, difficulty, and faithfulness of the underlying problems. Neglecting the fundamental importance of data has given rise to inaccuracy, bias, and fragility in real-world applications, and research is hindered by saturation across existing dataset benchmarks. In response, we present DataPerf, a community-led benchmark suite for evaluating ML datasets and data-centric algorithms. We aim to foster innovation in data-centric AI through competition, comparability, and reproducibility. We enable the ML community to iterate on datasets, instead of just architectures, and we provide an open, online platform with multiple rounds of challenges to support this iterative development. The first iteration of DataPerf contains five benchmarks covering a wide spectrum of data-centric techniques, tasks, and modalities in vision, speech, acquisition, debugging, and diffusion prompting, and we support hosting new contributed benchmarks from the community. The benchmarks, online evaluation platform, and baseline implementations are open source, and the MLCommons Association will maintain DataPerf to ensure long-term benefits to academia and industry.

Mark Zhao, Dhruv Choudhary, Devashish Tyagi et al. · stanford

We present RecD (Recommendation Deduplication), a suite of end-to-end infrastructure optimizations across the Deep Learning Recommendation Model (DLRM) training pipeline. RecD addresses immense storage, preprocessing, and training overheads caused by feature duplication inherent in industry-scale DLRM training datasets. Feature duplication arises because DLRM datasets are generated from interactions. While each user session can generate multiple training samples, many features' values do not change across these samples. We demonstrate how RecD exploits this property, end-to-end, across a deployed training pipeline. RecD optimizes data generation pipelines to decrease dataset storage and preprocessing resource demands and to maximize duplication within a training batch. RecD introduces a new tensor format, InverseKeyedJaggedTensors (IKJTs), to deduplicate feature values in each batch. We show how DLRM model architectures can leverage IKJTs to drastically increase training throughput. RecD improves the training and preprocessing throughput and storage efficiency by up to 2.48x, 1.79x, and 3.71x, respectively, in an industry-scale DLRM training system.

Geet Sethi, Pallab Bhattacharya, Dhruv Choudhary et al. · stanford

Sequence-based deep learning recommendation models (DLRMs) are an emerging class of DLRMs showing great improvements over their prior sum-pooling based counterparts at capturing users' long term interests. These improvements come at immense system cost however, with sequence-based DLRMs requiring substantial amounts of data to be dynamically materialized and communicated by each accelerator during a single iteration. To address this rapidly growing bottleneck, we present FlexShard, a new tiered sequence embedding table sharding algorithm which operates at a per-row granularity by exploiting the insight that not every row is equal. Through precise replication of embedding rows based on their underlying probability distribution, along with the introduction of a new sharding strategy adapted to the heterogeneous, skewed performance of real-world cluster network topologies, FlexShard is able to significantly reduce communication demand while using no additional memory compared to the prior state-of-the-art. When evaluated on production-scale sequence DLRMs, FlexShard was able to reduce overall global all-to-all communication traffic by over 85%, resulting in end-to-end training communication latency improvements of almost 6x over the prior state-of-the-art approach.

Kiwan Maeng, Haiyu Lu, Luca Melis et al.

Federated learning (FL) is an effective mechanism for data privacy in recommender systems by running machine learning model training on-device. While prior FL optimizations tackled the data and system heterogeneity challenges faced by FL, they assume the two are independent of each other. This fundamental assumption is not reflective of real-world, large-scale recommender systems -- data and system heterogeneity are tightly intertwined. This paper takes a data-driven approach to show the inter-dependence of data and system heterogeneity in real-world data and quantifies its impact on the overall model quality and fairness. We design a framework, RF^2, to model the inter-dependence and evaluate its impact on state-of-the-art model optimization techniques for federated recommendation tasks. We demonstrate that the impact on fairness can be severe under realistic heterogeneity scenarios, by up to 15.8--41x compared to a simple setup assumed in most (if not all) prior work. It means when realistic system-induced data heterogeneity is not properly modeled, the fairness impact of an optimization can be downplayed by up to 41x. The result shows that modeling realistic system-induced data heterogeneity is essential to achieving fair federated recommendation learning. We plan to open-source RF^2 to enable future design and evaluation of FL innovations.

Yejin Lee, Anna Sun, Basil Hosmer et al. · meta-ai, stanford

Generative artificial intelligence (AI) technology is revolutionizing the computing industry. Not only its applications have broadened to various sectors but also poses new system design and optimization opportunities. The technology is capable of understanding and responding in multiple modalities. However, the advanced capability currently comes with significant system resource demands. To sustainably scale generative AI capabilities to billions of users in the world, inference must be fast and efficient. This paper pinpoints key system design and optimization opportunities by characterizing a family of emerging multi-modal generation models on real systems. Auto-regressive token generation is a critical latency performance bottleneck, typically dominated by GPU idle time. In addition to memory-intensive attention across the generative AI models, linear operations constitute significant inference latency due to the feed forward networks in Transformer-based models. We demonstrate that state-of-the-art optimization levers, spanning from applications to system software and hardware, set a 3.88x better baseline.

Noveen Sachdeva, Mehak Preet Dhaliwal, Carole-Jean Wu et al.

We leverage the Neural Tangent Kernel and its equivalence to training infinitely-wide neural networks to devise $\infty$-AE: an autoencoder with infinitely-wide bottleneck layers. The outcome is a highly expressive yet simplistic recommendation model with a single hyper-parameter and a closed-form solution. Leveraging $\infty$-AE's simplicity, we also develop Distill-CF for synthesizing tiny, high-fidelity data summaries which distill the most important knowledge from the extremely large and sparse user-item interaction matrix for efficient and accurate subsequent data-usage like model training, inference, architecture search, etc. This takes a data-centric approach to recommendation, where we aim to improve the quality of logged user-feedback data for subsequent modeling, independent of the learning algorithm. We particularly utilize the concept of differentiable Gumbel-sampling to handle the inherent data heterogeneity, sparsity, and semi-structuredness, while being scalable to datasets with hundreds of millions of user-item interactions. Both of our proposed approaches significantly outperform their respective state-of-the-art and when used together, we observe 96-105% of $\infty$-AE's performance on the full dataset with as little as 0.1% of the original dataset size, leading us to explore the counter-intuitive question: Is more data what you need for better recommendation?

Newsha Ardalani, Carole-Jean Wu, Zeliang Chen et al.

Scale has been a major driving force in improving machine learning performance, and understanding scaling laws is essential for strategic planning for a sustainable model quality performance growth, long-term resource planning and developing efficient system infrastructures to support large-scale models. In this paper, we study empirical scaling laws for DLRM style recommendation models, in particular Click-Through Rate (CTR). We observe that model quality scales with power law plus constant in model size, data size and amount of compute used for training. We characterize scaling efficiency along three different resource dimensions, namely data, parameters and compute by comparing the different scaling schemes along these axes. We show that parameter scaling is out of steam for the model architecture under study, and until a higher-performing model architecture emerges, data scaling is the path forward. The key research questions addressed by this study include: Does a recommendation model scale sustainably as predicted by the scaling laws? Or are we far off from the scaling law predictions? What are the limits of scaling? What are the implications of the scaling laws on long-term hardware/system development?

Haiyang Huang, Newsha Ardalani, Anna Sun et al.

Mixture-of-Experts (MoE) models have gained popularity in achieving state-of-the-art performance in a wide range of tasks in computer vision and natural language processing. They effectively expand the model capacity while incurring a minimal increase in computation cost during training. However, deploying such models for inference is difficult due to their large size and complex communication pattern. In this work, we provide a characterization of two MoE workloads, namely Language Modeling (LM) and Machine Translation (MT) and identify their sources of inefficiencies at deployment. We propose three optimization techniques to mitigate sources of inefficiencies, namely (1) Dynamic gating, (2) Expert Buffering, and (3) Expert load balancing. We show that dynamic gating improves maximum throughput by 6.21-11.23$\times$ for LM, 5.75-10.98$\times$ for MT Encoder and 2.58-5.71$\times$ for MT Decoder. It also reduces memory usage by up to 1.36$\times$ for LM and up to 1.1$\times$ for MT. We further propose Expert Buffering, a new caching mechanism that only keeps hot, active experts in GPU memory while buffering the rest in CPU memory. This reduces static memory allocation by up to 1.47$\times$. We finally propose a load balancing methodology that provides additional scalability to the workload.

Samuel Hsia, Udit Gupta, Bilge Acun et al.

Deep learning recommendation systems serve personalized content under diverse tail-latency targets and input-query loads. In order to do so, state-of-the-art recommendation models rely on terabyte-scale embedding tables to learn user preferences over large bodies of contents. The reliance on a fixed embedding representation of embedding tables not only imposes significant memory capacity and bandwidth requirements but also limits the scope of compatible system solutions. This paper challenges the assumption of fixed embedding representations by showing how synergies between embedding representations and hardware platforms can lead to improvements in both algorithmic- and system performance. Based on our characterization of various embedding representations, we propose a hybrid embedding representation that achieves higher quality embeddings at the cost of increased memory and compute requirements. To address the system performance challenges of the hybrid representation, we propose MP-Rec -- a co-design technique that exploits heterogeneity and dynamic selection of embedding representations and underlying hardware platforms. On real system hardware, we demonstrate how matching custom accelerators, i.e., GPUs, TPUs, and IPUs, with compatible embedding representations can lead to 16.65x performance speedup. Additionally, in query-serving scenarios, MP-Rec achieves 2.49x and 3.76x higher correct prediction throughput and 0.19% and 0.22% better model quality on a CPU-GPU system for the Kaggle and Terabyte datasets, respectively.

Ashkan Yousefpour, Shen Guo, Ashish Shenoy et al.

The rapid progress of AI is fueled by increasingly large and computationally intensive machine learning models and datasets. As a consequence, the amount of compute used in training state-of-the-art models is exponentially increasing (doubling every 10 months between 2015 and 2022), resulting in a large carbon footprint. Federated Learning (FL) - a collaborative machine learning technique for training a centralized model using data of decentralized entities - can also be resource-intensive and have a significant carbon footprint, particularly when deployed at scale. Unlike centralized AI that can reliably tap into renewables at strategically placed data centers, cross-device FL may leverage as many as hundreds of millions of globally distributed end-user devices with diverse energy sources. Green AI is a novel and important research area where carbon footprint is regarded as an evaluation criterion for AI, alongside accuracy, convergence speed, and other metrics. In this paper, we propose the concept of Green FL, which involves optimizing FL parameters and making design choices to minimize carbon emissions consistent with competitive performance and training time. The contributions of this work are two-fold. First, we adopt a data-driven approach to quantify the carbon emissions of FL by directly measuring real-world at-scale FL tasks running on millions of phones. Second, we present challenges, guidelines, and lessons learned from studying the trade-off between energy efficiency, performance, and time-to-train in a production FL system. Our findings offer valuable insights into how FL can reduce its carbon footprint, and they provide a foundation for future research in the area of Green AI.

Meisam Hejazinia, Dzmitry Huba, Ilias Leontiadis et al.

Federated learning (FL) has emerged as an effective approach to address consumer privacy needs. FL has been successfully applied to certain machine learning tasks, such as training smart keyboard models and keyword spotting. Despite FL's initial success, many important deep learning use cases, such as ranking and recommendation tasks, have been limited from on-device learning. One of the key challenges faced by practical FL adoption for DL-based ranking and recommendation is the prohibitive resource requirements that cannot be satisfied by modern mobile systems. We propose Federated Ensemble Learning (FEL) as a solution to tackle the large memory requirement of deep learning ranking and recommendation tasks. FEL enables large-scale ranking and recommendation model training on-device by simultaneously training multiple model versions on disjoint clusters of client devices. FEL integrates the trained sub-models via an over-arch layer into an ensemble model that is hosted on the server. Our experiments demonstrate that FEL leads to 0.43-2.31% model quality improvement over traditional on-device federated learning - a significant improvement for ranking and recommendation system use cases.

Young Geun Kim, Carole-Jean Wu

Federated learning (FL) has emerged as a solution to deal with the risk of privacy leaks in machine learning training. This approach allows a variety of mobile devices to collaboratively train a machine learning model without sharing the raw on-device training data with the cloud. However, efficient edge deployment of FL is challenging because of the system/data heterogeneity and runtime variance. This paper optimizes the energy-efficiency of FL use cases while guaranteeing model convergence, by accounting for the aforementioned challenges. We propose FedGPO based on a reinforcement learning, which learns how to identify optimal global parameters (B, E, K) for each FL aggregation round adapting to the system/data heterogeneity and stochastic runtime variance. In our experiments, FedGPO improves the model convergence time by 2.4 times, and achieves 3.6 times higher energy efficiency over the baseline settings, respectively.

Lingjiao Chen, Bilge Acun, Newsha Ardalani et al.

As Machine Learning (ML) systems continue to grow, the demand for relevant and comprehensive datasets becomes imperative. There is limited study on the challenges of data acquisition due to ad-hoc processes and lack of consistent methodologies. We first present an investigation of current data marketplaces, revealing lack of platforms offering detailed information about datasets, transparent pricing, standardized data formats. With the objective of inciting participation from the data-centric AI community, we then introduce the DAM challenge, a benchmark to model the interaction between the data providers and acquirers. The benchmark was released as a part of DataPerf. Our evaluation of the submitted strategies underlines the need for effective data acquisition strategies in ML.

Karen Hambardzumyan, Nicolas Baldwin, Edan Toledo et al.

Existing research has identified three structural performance bottlenecks in AI research agents: (1) synchronous single-GPU execution constrains sample throughput, limiting the benefit of search; (2) a generalization gap where validation-based selection causes performance to degrade over extended search horizons; and (3) the limited capability of fixed, single-turn LLM operators imposes a ceiling on search performance. We introduce AIRA$_2$, which addresses these bottlenecks through three architectural choices: an asynchronous multi-GPU worker pool that increases experiment throughput linearly; a Hidden Consistent Evaluation protocol that delivers a reliable evaluation signal; and ReAct agents that dynamically scope their actions and debug interactively. On MLE-bench-30, AIRA$_2$ achieves a mean Percentile Rank of 71.8% at 24 hours - surpassing the previous best of 69.9% - and steadily improves to 76.0% at 72 hours. Ablation studies reveal that each component is necessary and that the "overfitting" reported in prior work was driven by evaluation noise rather than true data memorization.

Samuel Hsia, Alicia Golden, Bilge Acun et al.

Training and deploying large-scale machine learning models is time-consuming, requires significant distributed computing infrastructures, and incurs high operational costs. Our analysis, grounded in real-world large model training on datacenter-scale infrastructures, reveals that 14~32% of all GPU hours are spent on communication with no overlapping computation. To minimize this outstanding communication latency and other inherent at-scale inefficiencies, we introduce an agile performance modeling framework, MAD-Max. This framework is designed to optimize parallelization strategies and facilitate hardware-software co-design opportunities. Through the application of MAD-Max to a suite of real-world large-scale ML models on state-of-the-art GPU clusters, we showcase potential throughput enhancements of up to 2.24x for pre-training and up to 5.2x for inference scenarios, respectively.

Mostafa Elhoushi, Akshat Shrivastava, Diana Liskovich et al. · meta-ai

We present LayerSkip, an end-to-end solution to speed-up inference of large language models (LLMs). First, during training we apply layer dropout, with low dropout rates for earlier layers and higher dropout rates for later layers, and an early exit loss where all transformer layers share the same exit. Second, during inference, we show that this training recipe increases the accuracy of early exit at earlier layers, without adding any auxiliary layers or modules to the model. Third, we present a novel self-speculative decoding solution where we exit at early layers and verify and correct with remaining layers of the model. Our proposed self-speculative decoding approach has less memory footprint than other speculative decoding approaches and benefits from shared compute and activations of the draft and verification stages. We run experiments on different Llama model sizes on different types of training: pretraining from scratch, continual pretraining, finetuning on specific data domain, and finetuning on specific task. We implement our inference solution and show speedups of up to 2.16x on summarization for CNN/DM documents, 1.82x on coding, and 2.0x on TOPv2 semantic parsing task. We open source our code and checkpoints at https://github.com/facebookresearch/LayerSkip.

Minghao Li, Alicia Golden, Samuel Hsia et al.

The rapid scaling of large language model training requires distributing GPU resources across multiple data center buildings and regions. We refer to such paradigm as "scale-across" training. As infrastructure expands, the system design space becomes increasingly intricate, encompassing new model architectures, hardware heterogeneity, and evolving communication patterns. Drawing from Meta's production experience, we highlight the complexities of deploying training jobs across a few data centers housing hundreds of thousands of GPUs. To accelerate exploration of the large design space and to enable efficient training for frontier model development, we conduct in-depth characterization of three key design dimensions: parallelism placement, parallelism scheduling, and network layer technologies. We then propose ScaleAcross Explorer, an optimizer that considers the interplay of design dimensions and holistically optimizes scale-across training. Testbed experiments and simulations demonstrate up to 64.62% training speedups over production configuration and up to 37.59% training speedups over the state-of-the-art baseline across a wide range of design points.

Alicia Golden, Michael Kuchnik, Samuel Hsia et al.

Large model training beyond tens of thousands of GPUs is an uncharted territory. At such scales, disruptions to the training process are not a matter of if, but a matter of when -- a stochastic process degrading training productivity. Dynamic runtime variation will become increasingly more frequent as training scales up and as GPUs are operated in increasingly power-limited and thermally-stressed environments. At the 64,000+ GPU scale, we already observe 9% GPU time variability for frontier foundation model training. Motivated by our analysis and the large design space around performance variability, we present PRISM -- a performance modeling framework that captures the stochastic nature of large-scale distributed training. The core of PRISM is a statistical method that quantifies probabilistic guarantees on training time. Using PRISM, we explore the design and optimization space of distributed training, enabling principled, variability-aware decisions that improve performance and system efficiency at scale.

Meghana Madhyastha, Daniel Haziza, Jesse Cai et al.

Trainings of Large Language Models are generally bottlenecked by matrix multiplications. In the Transformer architecture, a large portion of these operations happens in the Feed Forward Network (FFN), and this portion increases for larger models, up to 50% of the total pretraining floating point operations. We show that we can leverage hardware-accelerated sparsity to accelerate all matrix multiplications in the FFN, with 2:4 sparsity for weights and v:n:m (Venom) sparsity for activations. Our recipe relies on sparse training steps to accelerate a large part of the pretraining, associated with regular dense training steps towards the end. Overall, models trained with this approach exhibit the same performance on our quality benchmarks, and can speed up training end-to-end by 1.4 to 1.7x. This approach is applicable to all NVIDIA GPUs starting with the A100 generation, and is orthogonal to common optimization techniques, such as, quantization, and can also be applied to mixture-of-experts model architectures.

Gang Liao, Hongsen Qin, Ying Wang et al.

Making deep learning recommendation model (DLRM) training and inference fast and efficient is important. However, this presents three key system challenges - model architecture diversity, kernel primitive diversity, and hardware generation and architecture heterogeneity. This paper presents KernelEvolve-an agentic kernel coding framework-to tackle heterogeneity at-scale for DLRM. KernelEvolve is designed to take kernel specifications as input and automate the process of kernel generation and optimization for recommendation model across heterogeneous hardware architectures. KernelEvolve does so by operating at multiple programming abstractions, from Triton and CuTe DSL to low-level hardware agnostic languages, spanning the full hardware-software optimization stack. The kernel optimization process is described as graph-based search with selection policy, universal operator, fitness function, and termination rule, dynamically adapts to runtime execution context through retrieval-augmented prompt synthesis. We designed, implemented, and deployed KernelEvolve to optimize a wide variety of production recommendation models across generations of NVIDIA and AMD GPUs, as well as Meta's AI accelerators. We validate KernelEvolve on the publicly-available KernelBench suite, achieving 100% pass rate on all 250 problems across three difficulty levels, and 160 PyTorch ATen operators across three heterogeneous hardware platforms, demonstrating 100% correctness. KernelEvolve reduces development time from weeks to hours and achieves substantial performance improvements over PyTorch baselines across diverse production use cases and for heterogeneous AI systems at-scale. Beyond performance efficiency improvements, KernelEvolve significantly mitigates the programmability barrier for new AI hardware by enabling automated kernel generation for in-house developed AI hardware.

Alberto Pepe, Chien-Yu Lin, Despoina Magka et al.

Toward recursive self-improvement, we investigate LLM agents autonomously designing foundation models beyond standard Transformers. We introduce a dual-framework approach: AIRA-Compose for high-level architecture search, and AIRA-Design for low-level mechanistic implementation. AIRA-Compose uses 11 agents to explore fundamental computational primitives under a 24-hour budget. Agents evaluate million-parameter candidates, extrapolating top designs to 350M, 1B, and 3B scales. This yields 14 architectures across two families: AIRAformers (Transformer-based) and AIRAhybrids (Transformer-Mamba). Pre-trained at 1B scale, these consistently outperform Llama 3.2 and Composer-found baselines. On downstream tasks, AIRAformer-D and AIRAhybrid-D improve accuracy by 2.4% and 3.8% over Llama 3.2. Furthermore, AIRA-Compose finds models with highly efficient scaling frontiers: AIRAformer-C scales 54% and 71% faster than Llama 3.2 and Composer's best Transformer, while AIRAhybrid-C outscales Nemotron-2 by 23% and Composer's best hybrid by 37%. AIRA-Design tasks 20 agents with writing novel attention mechanisms for long-range dependencies and high-performing training scripts. On the Long Range Arena benchmark, agent-designed architectures reach within 2.3% and 2.6% of human state-of-the-art on document matching and text classification. On the Autoresearch benchmark, Greedy Opus 4.5 achieves 0.968 validation bits-per-byte under a fixed time budget, surpassing the published minimum. Together, these frameworks show AI agents can autonomously discover architectures and algorithmic optimizations matching or surpassing hand-designed baselines. This establishes a powerful paradigm for discovering next-generation foundation models, marking a clear step toward recursive self-improvement.

Dong Wang, Yang Li, Ansong Ni et al.

Synthetic data has become increasingly important for training large language models, especially when real data is scarce, expensive, or privacy-sensitive. Many such generation tasks require coordinated multi-agent workflows, where specialized agents collaborate to produce data that is higher quality, more diverse, and structurally richer. However, existing frameworks for multi-agent synthesis often depend on a centralized orchestrator, creating scalability bottlenecks, or are hardcoded for specific domains, limiting flexibility. We present \textbf{Matrix}, a decentralized framework that represents both control and data flow as serialized messages passed through distributed queues. This peer-to-peer design eliminates the central orchestrator. Each task progresses independently through lightweight agents, while compute-intensive operations, such as LLM inference or containerized environments, are handled by distributed services. Built on Ray, Matrix scales to tens of thousands of concurrent agentic workflows and provides a modular, configurable design that enables easy adaptation to a wide range of data generation workflows. We evaluate Matrix across diverse synthesis scenarios, such as multi-agent collaborative dialogue, web-based reasoning data extraction, and tool-use trajectory generation in customer service environments. In all cases, Matrix achieves $2$--$15\times$ higher data generation throughput under identical hardware resources, without compromising output quality.

Irene Wang, Newsha Ardalani, Mostafa Elhoushi et al.

Machine learning solutions are rapidly adopted to enable a variety of key use cases, from conversational AI assistants to scientific discovery. This growing adoption is expected to increase the associated lifecycle carbon footprint, including both \emph{operational carbon} from training and inference and \emph{embodied carbon} from AI hardware manufacturing. We introduce \ourframework -- the first carbon-aware co-optimization framework for Transformer-based models and hardware accelerators. By integrating both operational and embodied carbon into early-stage design space exploration, \ourframework enables sustainability-driven model architecture and hardware accelerator co-design that reveals fundamentally different trade-offs than latency- or energy-centric approaches. Evaluated across a range of Transformer models, \ourframework consistently demonstrates the potential to reduce total carbon emissions -- by up to 30\% -- while maintaining accuracy and latency. We further highlight its extensibility through a focused case study on multi-modal models. Our results emphasize the need for holistic optimization methods that prioritize carbon efficiency without compromising model capability and execution time performance. The source code of \ourframework is available at {\small{\href{https://github.com/facebookresearch/CATransformers}{\texttt{https://github.com/facebookresearch/CATransformers}}}}.

Alicia Golden, Carole-Jean Wu, Gu-Yeon Wei et al.

The push for greater efficiency in AI computation has given rise to an array of accelerator architectures that increasingly challenge the GPU's long-standing dominance. In this work, we provide a quantitative view of this evolving landscape of AI accelerators, including the Cerebras CS-3, SambaNova SN-40, Groq, Gaudi, and TPUv5e platforms, and compare against both NVIDIA (A100, H100) and AMD (MI-300X) GPUs. We evaluate key trade-offs in latency, throughput, power consumption, and energy-efficiency across both (i) end-to-end workloads and (ii) benchmarks of individual computational primitives. Notably, we find the optimal hardware platform varies across batch size, sequence length, and model size, revealing a large underlying optimization space. Our analysis includes detailed power measurements across the prefill and decode phases of LLM inference, as well as quantification of the energy cost of communication. We additionally find that Cerebras, SambaNova, and Gaudi have 10-60% higher idle power than NVIDIA and AMD GPUs, emphasizing the importance of high utilization in order to realize promised efficiency gains. Finally, we assess programmability across platforms based on our experiments with real profiled workloads, comparing the compilation times and software stack maturity required to achieve promised performance.

Feiyang Kang, Michael Kuchnik, Karthik Padthe et al.

In post-training for reasoning Large Language Models (LLMs), the current state of practice trains LLMs in two independent stages: Supervised Fine-Tuning (SFT) and Reinforcement Learning with Verifiable Rewards (RLVR, shortened as ``RL'' below). In this work, we challenge whether high SFT scores translate to improved performance after RL. We provide extensive counter-examples where this is not true. We find high SFT scores can be biased toward simpler or more homogeneous data and are not reliably predictive of subsequent RL gains or scaled-up post-training effectiveness. In some cases, RL training on models with improved SFT performance could lead to substantially worse outcome compared to RL on the base model without SFT. We study alternative metrics and identify generalization loss on held-out reasoning examples and Pass@large k performance to provide strong proxies for the RL outcome. We trained hundreds of models up to 12B-parameter with SFT and RLVR via GRPO and ran extensive evaluations on 7 math benchmarks with up to 256 repetitions, spending $>$1M GPU hours. Experiments include models from Llama3, Mistral-Nemo, Qwen3 and multiple state-of-the-art SFT/RL datasets. Compared to directly predicting from pre-RL performance, prediction based on generalization loss and Pass@large k achieves substantial higher precision, improving $R^2$ coefficient and Spearman's rank correlation coefficient by up to 0.5 (2x). This provides strong utility for broad use cases. For example, in most experiments, we find SFT training on unique examples for a one epoch underperforms training on half examples for two epochs, either after SFT or SFT-then-RL; With the same SFT budget, training only on short examples may lead to better SFT performance, though, it often leads to worse outcome after RL compared to training on examples with varying lengths. Evaluation tool will be open-sourced.

Udit Gupta, Carole-Jean Wu, Xiaodong Wang et al.

The widespread application of deep learning has changed the landscape of computation in the data center. In particular, personalized recommendation for content ranking is now largely accomplished leveraging deep neural networks. However, despite the importance of these models and the amount of compute cycles they consume, relatively little research attention has been devoted to systems for recommendation. To facilitate research and to advance the understanding of these workloads, this paper presents a set of real-world, production-scale DNNs for personalized recommendation coupled with relevant performance metrics for evaluation. In addition to releasing a set of open-source workloads, we conduct in-depth analysis that underpins future system design and optimization for at-scale recommendation: Inference latency varies by 60% across three Intel server generations, batching and co-location of inferences can drastically improve latency-bounded throughput, and the diverse composition of recommendation models leads to different optimization strategies.

Mubashara Akhtar, Omar Benjelloun, Costanza Conforti et al.

Data is a critical resource for machine learning (ML), yet working with data remains a key friction point. This paper introduces Croissant, a metadata format for datasets that creates a shared representation across ML tools, frameworks, and platforms. Croissant makes datasets more discoverable, portable, and interoperable, thereby addressing significant challenges in ML data management. Croissant is already supported by several popular dataset repositories, spanning hundreds of thousands of datasets, enabling easy loading into the most commonly-used ML frameworks, regardless of where the data is stored. Our initial evaluation by human raters shows that Croissant metadata is readable, understandable, complete, yet concise.

Jhe-Yu Liou, Stephanie Forrest, Carole-Jean Wu

Parallel accelerators, such as GPUs, are key enablers for large-scale Machine Learning (ML) applications. However, ML model developers often lack detailed knowledge of the underlying system architectures, while system programmers usually do not have a high-level understanding of the ML model that runs on the specific system. To mitigate this gap between two relevant aspects of domain knowledge, this paper proposes GEVO-ML, a tool for automatically discovering optimization opportunities and tuning the performance of ML kernels, where the model and training/prediction processes are uniformly represented in a single intermediate language, the Multiple-Layer Intermediate Representation (MLIR). GEVO-ML uses multi-objective evolutionary search to find edits (mutations) to MLIR code that ultimately runs on GPUs, improving performance on desired criteria while retaining required functionality. We demonstrate GEVO-ML on two different ML workloads for both model training and prediction. GEVO-ML finds significant Pareto improvements for these models, achieving 90.43% performance improvement when model accuracy is relaxed by 2%, from 91.2% to 89.3%. For the training workloads, GEVO-ML finds a 4.88% improvement in model accuracy, from 91% to 96%, without sacrificing training or testing speed. Our analysis of key GEVO-ML mutations reveals diverse code modifications, while might be foreign to human developers, achieving similar effects with how human developers improve model design, for example, by changing learning rates or pruning non-essential layer parameters.

Bertie Vidgen, Adarsh Agrawal, Ahmed M. Ahmed et al. · deepmind, oxford

This paper introduces v0.5 of the AI Safety Benchmark, which has been created by the MLCommons AI Safety Working Group. The AI Safety Benchmark has been designed to assess the safety risks of AI systems that use chat-tuned language models. We introduce a principled approach to specifying and constructing the benchmark, which for v0.5 covers only a single use case (an adult chatting to a general-purpose assistant in English), and a limited set of personas (i.e., typical users, malicious users, and vulnerable users). We created a new taxonomy of 13 hazard categories, of which 7 have tests in the v0.5 benchmark. We plan to release version 1.0 of the AI Safety Benchmark by the end of 2024. The v1.0 benchmark will provide meaningful insights into the safety of AI systems. However, the v0.5 benchmark should not be used to assess the safety of AI systems. We have sought to fully document the limitations, flaws, and challenges of v0.5. This release of v0.5 of the AI Safety Benchmark includes (1) a principled approach to specifying and constructing the benchmark, which comprises use cases, types of systems under test (SUTs), language and context, personas, tests, and test items; (2) a taxonomy of 13 hazard categories with definitions and subcategories; (3) tests for seven of the hazard categories, each comprising a unique set of test items, i.e., prompts. There are 43,090 test items in total, which we created with templates; (4) a grading system for AI systems against the benchmark; (5) an openly available platform, and downloadable tool, called ModelBench that can be used to evaluate the safety of AI systems on the benchmark; (6) an example evaluation report which benchmarks the performance of over a dozen openly available chat-tuned language models; (7) a test specification for the benchmark.

Apostolos Kokolis, Michael Kuchnik, John Hoffman et al.

Reliability is a fundamental challenge in operating large-scale machine learning (ML) infrastructures, particularly as the scale of ML models and training clusters continues to grow. Despite decades of research on infrastructure failures, the impact of job failures across different scales remains unclear. This paper presents a view of managing two large, multi-tenant ML clusters, providing quantitative analysis, operational experience, and our own perspective in understanding and addressing reliability concerns at scale. Our analysis reveals that while large jobs are most vulnerable to failures, smaller jobs make up the majority of jobs in the clusters and should be incorporated into optimization objectives. We identify key workload properties, compare them across clusters, and demonstrate essential reliability requirements for pushing the boundaries of ML training at scale. We hereby introduce a taxonomy of failures and key reliability metrics, analyze 11 months of data from two state-of-the-art ML environments with 4 million jobs and over 150 million A100 GPU hours. Building on our data, we fit a failure model to project Mean Time to Failure for various GPU scales. We further propose a method to estimate a related metric, Effective Training Time Ratio, as a function of job parameters, and we use this model to gauge the efficacy of potential software mitigations at scale. Our work provides valuable insights and future research directions for improving the reliability of AI supercomputer clusters, emphasizing the need for flexible, workload-agnostic, and reliability-aware infrastructure, system software, and algorithms.

Yu Yang, Aaditya K. Singh, Mostafa Elhoushi et al.

Code datasets, often collected from diverse and uncontrolled sources such as GitHub, potentially suffer from quality issues, thereby affecting the performance and training efficiency of Large Language Models (LLMs) optimized for code generation. Previous studies demonstrated the benefit of using embedding spaces for data pruning, but they mainly focused on duplicate removal or increasing variety, and in other modalities, such as images. Our work focuses on using embeddings to identify and remove "low-quality" code data. First, we explore features of "low-quality" code in embedding space, through the use of synthetic corruptions. Armed with this knowledge, we devise novel pruning metrics that operate in embedding space to identify and remove low-quality entries in the Stack dataset. We demonstrate the benefits of this synthetic corruption informed pruning (SCIP) approach on the well-established HumanEval and MBPP benchmarks, outperforming existing embedding-based methods. Importantly, we achieve up to a 3% performance improvement over no pruning, thereby showing the promise of insights from synthetic corruptions for data pruning.

Saurabh Agarwal, Bilge Acun, Basil Hosmer et al.

Large Language Models (LLMs) with hundreds of billions of parameters have transformed the field of machine learning. However, serving these models at inference time is both compute and memory intensive, where a single request can require multiple GPUs and tens of Gigabytes of memory. Multi-Head Attention is one of the key components of LLMs, which can account for over 50% of LLMs memory and compute requirement. We observe that there is a high amount of redundancy across heads on which tokens they pay attention to. Based on this insight, we propose Clustered Head Attention (CHAI). CHAI combines heads with a high amount of correlation for self-attention at runtime, thus reducing both memory and compute. In our experiments, we show that CHAI is able to reduce the memory requirements for storing K,V cache by up to 21.4% and inference time latency by up to 1.73x without any fine-tuning required. CHAI achieves this with a maximum 3.2% deviation in accuracy across 3 different models (i.e. OPT-66B, LLAMA-7B, LLAMA-33B) and 5 different evaluation datasets.

Alicia Golden, Samuel Hsia, Fei Sun et al.

As the development of large-scale Generative AI models evolve beyond text (1D) generation to include image (2D) and video (3D) generation, processing spatial and temporal information presents unique challenges to quality, performance, and efficiency. We present the first work towards understanding this new system design space for multi-modal text-to-image (TTI) and text-to-video (TTV) generation models. Current model architecture designs are bifurcated into 2 categories: Diffusion- and Transformer-based models. Our systematic performance characterization on a suite of eight representative TTI/TTV models shows that after state-of-the-art optimization techniques such as Flash Attention are applied, Convolution accounts for up to 44% of execution time for Diffusion-based TTI models, while Linear layers consume up to 49% of execution time for Transformer-based models. We additionally observe that Diffusion-based TTI models resemble the Prefill stage of LLM inference, and benefit from 1.1-2.5x greater speedup from Flash Attention than Transformer-based TTI models that resemble the Decode phase. Since optimizations designed for LLMs do not map directly onto TTI/TTV models, we must conduct a thorough characterization of these workloads to gain insights for new optimization opportunities. In doing so, we define sequence length in the context of TTI/TTV models and observe sequence length can vary up to 4x in Diffusion model inference. We additionally observe temporal aspects of TTV workloads pose unique system bottlenecks, with Temporal Attention accounting for over 60% of total Attention time. Overall, our in-depth system performance characterization is a critical first step towards designing efficient and deployable systems for emerging TTI/TTV workloads.

Edan Toledo, Karen Hambardzumyan, Martin Josifoski et al. · meta-ai, oxford

AI research agents are demonstrating great potential to accelerate scientific progress by automating the design, implementation, and training of machine learning models. We focus on methods for improving agents' performance on MLE-bench, a challenging benchmark where agents compete in Kaggle competitions to solve real-world machine learning problems. We formalize AI research agents as search policies that navigate a space of candidate solutions, iteratively modifying them using operators. By designing and systematically varying different operator sets and search policies (Greedy, MCTS, Evolutionary), we show that their interplay is critical for achieving high performance. Our best pairing of search strategy and operator set achieves a state-of-the-art result on MLE-bench lite, increasing the success rate of achieving a Kaggle medal from 39.6% to 47.7%. Our investigation underscores the importance of jointly considering the search strategy, operator design, and evaluation methodology in advancing automated machine learning.

Alicia Golden, Samuel Hsia, Fei Sun et al.

Training large-scale machine learning models poses distinct system challenges, given both the size and complexity of today's workloads. Recently, many organizations training state-of-the-art Generative AI models have reported cases of instability during training, often taking the form of loss spikes. Numeric deviation has emerged as a potential cause of this training instability, although quantifying this is especially challenging given the costly nature of training runs. In this work, we develop a principled approach to understanding the effects of numeric deviation, and construct proxies to put observations into context when downstream effects are difficult to quantify. As a case study, we apply this framework to analyze the widely-adopted Flash Attention optimization. We find that Flash Attention sees roughly an order of magnitude more numeric deviation as compared to Baseline Attention at BF16 when measured during an isolated forward pass. We then use a data-driven analysis based on the Wasserstein Distance to provide upper bounds on how this numeric deviation impacts model weights during training, finding that the numerical deviation present in Flash Attention is 2-5 times less significant than low-precision training.

Sangmin Bae, Bilge Acun, Haroun Habeeb et al.

Recent progress in large language models demonstrates that hybrid architectures--combining self-attention mechanisms with structured state space models like Mamba--can achieve a compelling balance between modeling quality and computational efficiency, particularly for long-context tasks. While these hybrid models show promising performance, systematic comparisons of hybridization strategies and analyses on the key factors behind their effectiveness have not been clearly shared to the community. In this work, we present a holistic evaluation of hybrid architectures based on inter-layer (sequential) or intra-layer (parallel) fusion. We evaluate these designs from a variety of perspectives: language modeling performance, long-context capabilities, scaling analysis, and training and inference efficiency. By investigating the core characteristics of their computational primitive, we identify the most critical elements for each hybridization strategy and further propose optimal design recipes for both hybrid models. Our comprehensive analysis provides practical guidance and valuable insights for developing hybrid language models, facilitating the optimization of architectural configurations.

Gyudong Kim, Mehdi Ghasemi, Soroush Heidari et al.

Federated Learning (FL) is a practical approach to train deep learning models collaboratively across user-end devices, protecting user privacy by retaining raw data on-device. In FL, participating user-end devices are highly fragmented in terms of hardware and software configurations. Such fragmentation introduces a new type of data heterogeneity in FL, namely \textit{system-induced data heterogeneity}, as each device generates distinct data depending on its hardware and software configurations. In this paper, we first characterize the impact of system-induced data heterogeneity on FL model performance. We collect a dataset using heterogeneous devices with variations across vendors and performance tiers. By using this dataset, we demonstrate that \textit{system-induced data heterogeneity} negatively impacts accuracy, and deteriorates fairness and domain generalization problems in FL. To address these challenges, we propose HeteroSwitch, which adaptively adopts generalization techniques (i.e., ISP transformation and SWAD) depending on the level of bias caused by varying HW and SW configurations. In our evaluation with a realistic FL dataset (FLAIR), HeteroSwitch reduces the variance of averaged precision by 6.3\% across device types.

Bilge Acun, Prasoon Sinha, Newsha Ardalani et al.

Hybrid model architectures that combine computational primitives (e.g., Attention, MLP) in different ratios have shown promising performance beyond Transformers. Some studies have shown that different interleavings of primitives can affect model quality as well. However, prior works explore the hybrid model architecture design space manually. Due to the large design space and training costs, discovering hybrid models that combine key computational primitives for pre-training is challenging. In this work, we take a principled approach in designing a modular hybrid model architecture search framework -- Composer. Composer explores model architectures at a small scale and extrapolates the top-performing model architectures to a larger scale using our proposed scaling strategies. Using Composer, we discover new hybrid LLM architectures that outperform Llama 3.2. Compared to Llama 3.2 and previous state-of-the-art baselines, the new model architectures consistently reduce validation loss at parameter scales of 350M-3B and improve evaluation accuracy on the downstream tasks by up to 2.8-8.3% (1.1-3.1% on average) while improving both training and inference efficiency.

Feiyang Kang, Newsha Ardalani, Michael Kuchnik et al. · meta-ai, mit

Training data plays a crucial role in Large Language Models (LLM) scaling, yet high quality data is of limited supply. Synthetic data techniques offer a potential path toward sidestepping these limitations. We conduct a large-scale empirical investigation (>1000 LLMs with >100k GPU hours) using a unified protocol and scaling laws, comparing natural web data, diverse synthetic types (rephrased text, generated textbooks), and mixtures of natural and synthetic data. Specifically, we found pre-training on rephrased synthetic data \textit{alone} is not faster than pre-training on natural web texts; while pre-training on 1/3 rephrased synthetic data mixed with 2/3 natural web texts can speed up 5-10x (to reach the same validation loss) at larger data budgets. Pre-training on textbook-style synthetic data \textit{alone} results in notably higher loss on many downstream domains especially at small data budgets. "Good" ratios of synthetic data in training data mixtures depend on the model size and data budget, empirically converging to ~30% for rephrased synthetic data. Larger generator models do not necessarily yield better pre-training data than ~8B-param models. These results contribute mixed evidence on "model collapse" during large-scale single-round (n=1) model training on synthetic data--training on rephrased synthetic data shows no degradation in performance in foreseeable scales whereas training on mixtures of textbook-style pure-generated synthetic data shows patterns predicted by "model collapse". Our work demystifies synthetic data in pre-training, validates its conditional benefits, and offers practical guidance.

Carole-Jean Wu, Bilge Acun, Ramya Raghavendra et al.

Barroso's seminal contributions in energy-proportional warehouse-scale computing launched an era where modern datacenters have become more energy efficient and cost effective than ever before. At the same time, modern AI applications have driven ever-increasing demands in computing, highlighting the importance of optimizing efficiency across the entire deep learning model development cycle. This paper characterizes the carbon impact of AI, including both operational carbon emissions from training and inference as well as embodied carbon emissions from datacenter construction and hardware manufacturing. We highlight key efficiency optimization opportunities for cutting-edge AI technologies, from deep learning recommendation models to multi-modal generative AI tasks. To scale AI sustainably, we must also go beyond efficiency and optimize across the life cycle of computing infrastructures, from hardware manufacturing to datacenter operations and end-of-life processing for the hardware.

John Nguyen, Sid Wang, Ke Li et al.

Fine-tuning large-scale Transformers has led to the explosion of many AI applications across Natural Language Processing and Computer Vision tasks. However, fine-tuning all pre-trained model parameters becomes impractical as the model size and number of tasks increase. Parameter-efficient transfer learning (PETL) methods aim to address these challenges. While effective in reducing the number of trainable parameters, PETL methods still require significant energy and computational resources to fine-tune. In this paper, we introduce \textbf{RE}current \textbf{AD}aption (READ) -- a lightweight and memory-efficient fine-tuning method -- to overcome the limitations of the current PETL approaches. Specifically, READ inserts a small RNN network alongside the backbone model so that the model does not have to back-propagate through the large backbone network. Through comprehensive empirical evaluation of the GLUE benchmark, we demonstrate READ can achieve a $56\%$ reduction in the training memory consumption and an $84\%$ reduction in the GPU energy usage while retraining high model quality compared to full-tuning. Additionally, the model size of READ does not grow with the backbone model size, making it a highly scalable solution for fine-tuning large Transformers.

Geet Sethi, Bilge Acun, Niket Agarwal et al.

We propose RecShard, a fine-grained embedding table (EMB) partitioning and placement technique for deep learning recommendation models (DLRMs). RecShard is designed based on two key observations. First, not all EMBs are equal, nor all rows within an EMB are equal in terms of access patterns. EMBs exhibit distinct memory characteristics, providing performance optimization opportunities for intelligent EMB partitioning and placement across a tiered memory hierarchy. Second, in modern DLRMs, EMBs function as hash tables. As a result, EMBs display interesting phenomena, such as the birthday paradox, leaving EMBs severely under-utilized. RecShard determines an optimal EMB sharding strategy for a set of EMBs based on training data distributions and model characteristics, along with the bandwidth characteristics of the underlying tiered memory hierarchy. In doing so, RecShard achieves over 6 times higher EMB training throughput on average for capacity constrained DLRMs. The throughput increase comes from improved EMB load balance by over 12 times and from the reduced access to the slower memory by over 87 times.

Noveen Sachdeva, Carole-Jean Wu, Julian McAuley

We study the practical consequences of dataset sampling strategies on the ranking performance of recommendation algorithms. Recommender systems are generally trained and evaluated on samples of larger datasets. Samples are often taken in a naive or ad-hoc fashion: e.g. by sampling a dataset randomly or by selecting users or items with many interactions. As we demonstrate, commonly-used data sampling schemes can have significant consequences on algorithm performance. Following this observation, this paper makes three main contributions: (1) characterizing the effect of sampling on algorithm performance, in terms of algorithm and dataset characteristics (e.g. sparsity characteristics, sequential dynamics, etc.); (2) designing SVP-CF, which is a data-specific sampling strategy, that aims to preserve the relative performance of models after sampling, and is especially suited to long-tailed interaction data; and (3) developing an oracle, Data-Genie, which can suggest the sampling scheme that is most likely to preserve model performance for a given dataset. The main benefit of Data-Genie is that it will allow recommender system practitioners to quickly prototype and compare various approaches, while remaining confident that algorithm performance will be preserved, once the algorithm is retrained and deployed on the complete data. Detailed experiments show that using Data-Genie, we can discard upto 5x more data than any sampling strategy with the same level of performance.

Dzmitry Huba, John Nguyen, Kshitiz Malik et al.

Cross-device Federated Learning (FL) is a distributed learning paradigm with several challenges that differentiate it from traditional distributed learning, variability in the system characteristics on each device, and millions of clients coordinating with a central server being primary ones. Most FL systems described in the literature are synchronous - they perform a synchronized aggregation of model updates from individual clients. Scaling synchronous FL is challenging since increasing the number of clients training in parallel leads to diminishing returns in training speed, analogous to large-batch training. Moreover, stragglers hinder synchronous FL training. In this work, we outline a production asynchronous FL system design. Our work tackles the aforementioned issues, sketches of some of the system design challenges and their solutions, and touches upon principles that emerged from building a production FL system for millions of clients. Empirically, we demonstrate that asynchronous FL converges faster than synchronous FL when training across nearly one hundred million devices. In particular, in high concurrency settings, asynchronous FL is 5x faster and has nearly 8x less communication overhead than synchronous FL.

Carole-Jean Wu, Ramya Raghavendra, Udit Gupta et al.

This paper explores the environmental impact of the super-linear growth trends for AI from a holistic perspective, spanning Data, Algorithms, and System Hardware. We characterize the carbon footprint of AI computing by examining the model development cycle across industry-scale machine learning use cases and, at the same time, considering the life cycle of system hardware. Taking a step further, we capture the operational and manufacturing carbon footprint of AI computing and present an end-to-end analysis for what and how hardware-software design and at-scale optimization can help reduce the overall carbon footprint of AI. Based on the industry experience and lessons learned, we share the key challenges and chart out important development directions across the many dimensions of AI. We hope the key messages and insights presented in this paper can inspire the community to advance the field of AI in an environmentally-responsible manner.

Mark Zhao, Niket Agarwal, Aarti Basant et al.

Datacenter-scale AI training clusters consisting of thousands of domain-specific accelerators (DSA) are used to train increasingly-complex deep learning models. These clusters rely on a data storage and ingestion (DSI) pipeline, responsible for storing exabytes of training data and serving it at tens of terabytes per second. As DSAs continue to push training efficiency and throughput, the DSI pipeline is becoming the dominating factor that constrains the overall training performance and capacity. Innovations that improve the efficiency and performance of DSI systems and hardware are urgent, demanding a deep understanding of DSI characteristics and infrastructure at scale. This paper presents Meta's end-to-end DSI pipeline, composed of a central data warehouse built on distributed storage and a Data PreProcessing Service that scales to eliminate data stalls. We characterize how hundreds of models are collaboratively trained across geo-distributed datacenters via diverse and continuous training jobs. These training jobs read and heavily filter massive and evolving datasets, resulting in popular features and samples used across training jobs. We measure the intense network, memory, and compute resources required by each training job to preprocess samples during training. Finally, we synthesize key takeaways based on our production infrastructure characterization. These include identifying hardware bottlenecks, discussing opportunities for heterogeneous DSI hardware, motivating research in datacenter scheduling and benchmark datasets, and assimilating lessons learned in optimizing DSI infrastructure.

Young Geun Kim, Carole-Jean Wu

Federated learning enables a cluster of decentralized mobile devices at the edge to collaboratively train a shared machine learning model, while keeping all the raw training samples on device. This decentralized training approach is demonstrated as a practical solution to mitigate the risk of privacy leakage. However, enabling efficient FL deployment at the edge is challenging because of non-IID training data distribution, wide system heterogeneity and stochastic-varying runtime effects in the field. This paper jointly optimizes time-to-convergence and energy efficiency of state-of-the-art FL use cases by taking into account the stochastic nature of edge execution. We propose AutoFL by tailor-designing a reinforcement learning algorithm that learns and determines which K participant devices and per-device execution targets for each FL model aggregation round in the presence of stochastic runtime variance, system and data heterogeneity. By considering the unique characteristics of FL edge deployment judiciously, AutoFL achieves 3.6 times faster model convergence time and 4.7 and 5.2 times higher energy efficiency for local clients and globally over the cluster of K participants, respectively.

Noveen Sachdeva, Carole-Jean Wu, Julian McAuley

We study the practical consequences of dataset sampling strategies on the performance of recommendation algorithms. Recommender systems are generally trained and evaluated on samples of larger datasets. Samples are often taken in a naive or ad-hoc fashion: e.g. by sampling a dataset randomly or by selecting users or items with many interactions. As we demonstrate, commonly-used data sampling schemes can have significant consequences on algorithm performance -- masking performance deficiencies in algorithms or altering the relative performance of algorithms, as compared to models trained on the complete dataset. Following this observation, this paper makes the following main contributions: (1) characterizing the effect of sampling on algorithm performance, in terms of algorithm and dataset characteristics (e.g. sparsity characteristics, sequential dynamics, etc.); and (2) designing SVP-CF, which is a data-specific sampling strategy, that aims to preserve the relative performance of models after sampling, and is especially suited to long-tail interaction data. Detailed experiments show that SVP-CF is more accurate than commonly used sampling schemes in retaining the relative ranking of different recommendation algorithms.

Zhaoxia, Deng, Jongsoo Park et al.

Tremendous success of machine learning (ML) and the unabated growth in ML model complexity motivated many ML-specific designs in both CPU and accelerator architectures to speed up the model inference. While these architectures are diverse, highly optimized low-precision arithmetic is a component shared by most. Impressive compute throughputs are indeed often exhibited by these architectures on benchmark ML models. Nevertheless, production models such as recommendation systems important to Facebook's personalization services are demanding and complex: These systems must serve billions of users per month responsively with low latency while maintaining high prediction accuracy, notwithstanding computations with many tens of billions parameters per inference. Do these low-precision architectures work well with our production recommendation systems? They do. But not without significant effort. We share in this paper our search strategies to adapt reference recommendation models to low-precision hardware, our optimization of low-precision compute kernels, and the design and development of tool chain so as to maintain our models' accuracy throughout their lifespan during which topic trends and users' interests inevitably evolve. Practicing these low-precision technologies helped us save datacenter capacities while deploying models with up to 5X complexity that would otherwise not be deployed on traditional general-purpose CPUs. We believe these lessons from the trenches promote better co-design between hardware architecture and software engineering and advance the state of the art of ML in industry.

Udit Gupta, Samuel Hsia, Jeff Zhang et al.

Deep learning recommendation systems must provide high quality, personalized content under strict tail-latency targets and high system loads. This paper presents RecPipe, a system to jointly optimize recommendation quality and inference performance. Central to RecPipe is decomposing recommendation models into multi-stage pipelines to maintain quality while reducing compute complexity and exposing distinct parallelism opportunities. RecPipe implements an inference scheduler to map multi-stage recommendation engines onto commodity, heterogeneous platforms (e.g., CPUs, GPUs).While the hardware-aware scheduling improves ranking efficiency, the commodity platforms suffer from many limitations requiring specialized hardware. Thus, we design RecPipeAccel (RPAccel), a custom accelerator that jointly optimizes quality, tail-latency, and system throughput. RPAc-cel is designed specifically to exploit the distinct design space opened via RecPipe. In particular, RPAccel processes queries in sub-batches to pipeline recommendation stages, implements dual static and dynamic embedding caches, a set of top-k filtering units, and a reconfigurable systolic array. Com-pared to prior-art and at iso-quality, we demonstrate that RPAccel improves latency and throughput by 3x and 6x.