In distributed machine learning (DML), the network performance between machines significantly impacts the speed of iterative training. In this paper we propose BML, a new gradient synchronization algorithm with higher network performance and lower network cost than the current practice. BML runs on BCube network, instead of using the traditional Fat-Tree topology. BML algorithm is designed in such a way that, compared to the parameter server (PS) algorithm on a Fat-Tree network connecting the same number of server machines, BML achieves theoretically 1/k of the gradient synchronization time, with k/5 of switches (the typical number of k is 2 4). Experiments of LeNet-5 and VGG-19 benchmarks on a testbed with 9 dual-GPU servers show that, BML reduces the job completion time of DML training by up to 56.4%.

Transformer-based large language models (LLMs) demonstrate impressive potential in various practical applications. However, long context inference poses a significant challenge due to the enormous memory requirements of the key-value (KV) cache, which can scale to multiple gigabytes as sequence length and batch size increase. In this paper, we present KVComp, a generic and efficient KV cache management framework optimized for long-text generation that synergistically works with both latency-critical and throughput-critical inference systems. KVComp employs novel lossy compression techniques specifically designed for KV cache data characteristics, featuring careful co-design of compression algorithms and system architecture. Our approach maintains compatibility with the growing nature of KV cache while preserving high computational efficiency. Experimental results show that KVComp achieves on average 47\% and up to 83\% higher memory reduction rate compared to existing methods with little/no model accuracy degradation. Furthermore, KVComp achieves extremely high execution throughput, effectively reducing decompression overhead and, in some cases, even accelerating the matrix-vector multiplication operation and outperform cuBLAS-based attention kernels with less data movement.

Despite increasing discussions on open-source Artificial Intelligence (AI), existing research lacks a discussion on the transparency and accessibility of state-of-the-art (SoTA) Large Language Models (LLMs). The Open Source Initiative (OSI) has recently released its first formal definition of open-source software. This definition, when combined with standard dictionary definitions and the sparse published literature, provide an initial framework to support broader accessibility to AI models such as LLMs, but more work is essential to capture the unique dynamics of openness in AI. In addition, concerns about open-washing, where models claim openness but lack full transparency, has been raised, which limits the reproducibility, bias mitigation, and domain adaptation of these models. In this context, our study critically analyzes SoTA LLMs from the last five years, including ChatGPT, DeepSeek, LLaMA, and others, to assess their adherence to transparency standards and the implications of partial openness. Specifically, we examine transparency and accessibility from two perspectives: open-source vs. open-weight models. Our findings reveal that while some models are labeled as open-source, this does not necessarily mean they are fully open-sourced. Even in the best cases, open-source models often do not report model training data, and code as well as key metrics, such as weight accessibility, and carbon emissions. To the best of our knowledge, this is the first study that systematically examines the transparency and accessibility of over 100 different SoTA LLMs through the dual lens of open-source and open-weight models. The findings open avenues for further research and call for responsible and sustainable AI practices to ensure greater transparency, accountability, and ethical deployment of these models.(DeepSeek transparency, ChatGPT accessibility, open source, DeepSeek open source)

Building a general-purpose unified model that can solve diverse tasks in different modalities while maintaining high performance is a long-standing challenge in the machine learning research community. A conventional approach in this direction is building models with task-specialized heads on top of a shared architectural backbone -- but such models require expert knowledge to design a specialized head for each task, and their lack of parameter-sharing for new tasks limits their transfer-learning capabilities. In the new paper Unified-IO: A Unified Model for Vision, Language, and Multi-Modal Tasks, a research team from the Allen Institute for AI and the University of Washington introduces UNIFIED-IO, a neural model with no task- or modality-specific branches that achieves competitive performance across a wide variety of computer vision (CV), natural language processing (NLP), and multi-modal benchmark tasks without fine-tuning. The researchers set out to build a unified neural architecture that ML practitioners with little or no knowledge of the underlying machinery could use to efficiently and effectively train their models for new NLP and CV tasks. For models to support a variety of modalities (images, language, boxes, binary masks, segmentation, etc.), they must represent all modalities in a shared space.

Artificial intelligence and machine learning (AI and ML) are key technologies that help organizations develop new ways to increase sales, reduce costs, streamline business processes, and understand their customers better. AWS helps customers accelerate their AI/ML adoption by delivering powerful compute, high-speed networking, and scalable high-performance storage options on demand for any machine learning project. This lowers the barrier to entry for organizations looking to adopt the cloud to scale their ML applications. Developers and data scientists are pushing the boundaries of technology and increasingly adopting deep learning, which is a type of machine learning based on neural network algorithms. These deep learning models are larger and more sophisticated resulting in rising costs to run underlying infrastructure to train and deploy these models.

Deep learning recommendation models (DLRMs) are used across many business-critical services at Facebook and are the single largest AI application in terms of infrastructure demand in its data-centers. In this paper we discuss the SW/HW co-designed solution for high-performance distributed training of large-scale DLRMs. We introduce a high-performance scalable software stack based on PyTorch and pair it with the new evolution of Zion platform, namely ZionEX. We demonstrate the capability to train very large DLRMs with up to 12 Trillion parameters and show that we can attain 40X speedup in terms of time to solution over previous systems. We achieve this by (i) designing the ZionEX platform with dedicated scale-out network, provisioned with high bandwidth, optimal topology and efficient transport (ii) implementing an optimized PyTorch-based training stack supporting both model and data parallelism (iii) developing sharding algorithms capable of hierarchical partitioning of the embedding tables along row, column dimensions and load balancing them across multiple workers; (iv) adding high-performance core operators while retaining flexibility to support optimizers with fully deterministic updates (v) leveraging reduced precision communications, multi-level memory hierarchy (HBM+DDR+SSD) and pipelining. Furthermore, we develop and briefly comment on distributed data ingestion and other supporting services that are required for the robust and efficient end-to-end training in production environments.

In distributed machine learning (DML), the network performance between machines significantly impacts the speed of iterative training. In this paper we propose BML, a new gradient synchronization algorithm with higher network performance and lower network cost than the current practice. BML runs on BCube network, instead of using the traditional Fat-Tree topology. BML algorithm is designed in such a way that, compared to the parameter server (PS) algorithm on a Fat-Tree network connecting the same number of server machines, BML achieves theoretically 1/k of the gradient synchronization time, with k/5 of switches (the typical number of k is 2 4). Experiments of LeNet-5 and VGG-19 benchmarks on a testbed with 9 dual-GPU servers show that, BML reduces the job completion time of DML training by up to 56.4%.