Many modern machine learning (ML) methods rely on embedding models to learn vector representations (embeddings) for a set of entities (embedding tables). As increasingly diverse ML applications utilize embedding models and embedding tables continue to grow in size and number, there has been a surge in the ad-hoc development of specialized frameworks targeted to train large embedding models for specific tasks. Although the scalability issues that arise in different embedding model training tasks are similar, each of these frameworks independently reinvents and customizes storage components for specific tasks, leading to substantial duplicated engineering efforts in both development and deployment. This paper presents MLKV, an efficient, extensible, and reusable data storage framework designed to address the scalability challenges in embedding model training, specifically data stall and staleness. MLKV augments disk-based key-value storage by democratizing optimizations that were previously exclusive to individual specialized frameworks and provides easy-to-use interfaces for embedding model training tasks. Extensive experiments on open-source workloads, as well as applications in eBay's payment transaction risk detection and seller payment risk detection, show that MLKV outperforms offloading strategies built on top of industrial-strength key-value stores by 1.6-12.6 ×. MLKV is open-source at https://github.com/llm-db/MLKV.

Methods for carefully selecting or generating a small set of training data to learn from, i.e., data pruning, coreset selection, and dataset distillation, have been shown to be effective in reducing the ever-increasing cost of training neural networks. Behind this success are rigorously designed, yet expensive, strategies for identifying the most informative training examples out of large datasets. In this work, we revisit these methods to understand if the additional computational costs associated with such strategies are justified from the perspective of time-to-accuracy, which has become a critical efficiency measure of deep neural network training over large datasets. Surprisingly, we find that many of the recently proposed methods underperform what we call Repeated Sampling of Random Subsets (RSRS or RS2), a powerful yet overlooked extension of the standard random baseline that learns from repeatedly sampled data throughout training instead of a fixed random subset. We test RS2 against thirty-two state-of-the-art data pruning and distillation methods across four datasets including ImageNet. Our results demonstrate that RS2 significantly reduces time-to-accuracy, particularly in practical regimes where accuracy, but not runtime, is similar to that of training on full dataset. For example, when training ResNet-18 on ImageNet, with 10\% of the dataset each epoch RS2 reaches an accuracy of 66\% versus 69\% when training with the full dataset. The best competing method achieves only 55\% while training 1.6$\times$ slower than RS2. Beyond the above meta-study, we discuss the theoretical properties of RS2 such as its convergence rate and generalization error. Our primary goal is to highlight that future works that aim to minimize total training cost by using subset selection, need to consider 1) the total computation cost (including preparing the subset) and 2) should aim to outperform a simple extension of random sampling (i.e., RS2).

A Retrieval-Augmented Language Model (RALM) combines a large language model (LLM) with a vector database to retrieve context-specific knowledge during text generation. This strategy facilitates impressive generation quality even with smaller models, thus reducing computational demands by orders of magnitude. To serve RALMs efficiently and flexibly, we propose Chameleon , a heterogeneous accelerator system integrating both LLM and vector search accelerators in a disaggregated architecture. The heterogeneity ensures efficient serving for both inference and retrieval, while the disaggregation allows independent scaling of LLM and vector search accelerators to fulfill diverse RALM requirements. Our Chameleon prototype implements vector search accelerators on FPGAs and assigns LLM inference to GPUs, with CPUs as cluster coordinators. Evaluated on various RALMs, Chameleon exhibits up to 2.16× reduction in latency and 3.18× speedup in throughput compared to the hybrid CPU-GPU architecture. The promising results pave the way for adopting heterogeneous accelerators for not only LLM inference but also vector search in future RALM systems.

We study training of Graph Neural Networks (GNNs) for large-scale graphs. We revisit the premise of using distributed training for billion-scale graphs and show that for graphs that fit in main memory or the SSD of a single machine, out-of-core pipelined training with a single GPU can outperform state-of-the-art (SoTA) multi-GPU solutions. We introduce MariusGNN, the first system that utilizes the entire storage hierarchy---including disk---for GNN training. MariusGNN introduces a series of data organization and algorithmic contributions that 1) minimize the end-to-end time required for training and 2) ensure that models learned with disk-based training exhibit accuracy similar to those fully trained in memory. We evaluate MariusGNN against SoTA systems for learning GNN models and find that single-GPU training in MariusGNN achieves the same level of accuracy up to 8× faster than multi-GPU training in these systems, thus, introducing an order of magnitude monetary cost reduction. MariusGNN is open-sourced at www.marius-project.org.

Graph embeddings have emerged as the de facto representation for modern machine learning over graph data structures. The goal of graph embedding models is to convert high-dimensional sparse graphs into low-dimensional, dense and continuous vector spaces that preserve the graph structure properties. However, learning a graph embedding model is a resource intensive process, and existing solutions rely on expensive distributed computation to scale training to instances that do not fit in GPU memory. This demonstration showcases Marius: a new open-source engine for learning graph embedding models over billion-edge graphs on a single machine. Marius is built around a recently-introduced architecture for machine learning over graphs that utilizes pipelining and a novel data replacement policy to maximize GPU utilization and exploit the entire memory hierarchy (including disk, CPU, and GPU memory) to scale to large instances. The audience will experience how to develop, train, and deploy graph embedding models using Marius' configuration-driven programming model. Moreover, the audience will have the opportunity to explore Marius' deployments on applications including link-prediction on WikiKG90M and reasoning queries on a paleobiology knowledge graph. Marius is available as open source software at https://marius-project.org.