Da Liu

Mingjun Zhang, Xiaohe Hu, Menghao Zhang et al.

Large-scale LLM training requires collective communication libraries to exchange data among distributed GPUs. As a company dedicated to building and operating large-scale GPU training clusters, we encounter several practical limitations of NCCL in production, including 1) SM competition between computation and communication, 2) expensive restart costs under link failures, and 3) insufficient observability of transient collective communication anomalies. To address these challenges, we propose VCCL, an efficient, reliable, and observable collective communication library in large-scale GPU training clusters. VCCL removes SM-consuming P2P kernels by moving intra-node data movement and stream dependency enforcement to CPU threads and GPU copy engines. VCCL also introduces a primary-backup QP mechanism to tolerate frequent NIC port failures, and designs a window-based monitor to observe network anomalies at O(μs) level. We opensource VCCL and deploy it in production training clusters for several months. Compared with NCCL, VCCL improves training throughput by up to 5.28% and reduces massive GPU resource wastage through runtime fault tolerance and finegrained monitor. We also share experience and lessons we learned during the deployment of VCCL in large-scale clusters.

Hengjie Yu, Dan Luo, Sam F. Y. Li et al.

Crops are constantly challenged by different environmental conditions. Seed treatment by nanomaterials is a cost-effective and environmentally-friendly solution for environmental stress mitigation in crop plants. Here, 56 seed nanopriming treatments are used to alleviate environmental stresses in maize. Seven selected nanopriming treatments significantly increase the stress resistance index (SRI) by 13.9% and 12.6% under salinity stress and combined heat-drought stress, respectively. Metabolomics data reveals that ZnO nanopriming treatment, with the highest SRI value, mainly regulates the pathways of amino acid metabolism, secondary metabolite synthesis, carbohydrate metabolism, and translation. Understanding the mechanism of seed nanopriming is still difficult due to the variety of nanomaterials and the complexity of interactions between nanomaterials and plants. Using the nanopriming data, we present an interpretable structure-activity relationship (ISAR) approach based on interpretable machine learning for predicting and understanding its stress mitigation effects. The post hoc and model-based interpretation approaches of machine learning are combined to provide complementary benefits and give researchers or policymakers more illuminating or trustworthy results. The concentration, size, and zeta potential of nanoparticles are identified as dominant factors for correlating root dry weight under salinity stress, and their effects and interactions are explained. Additionally, a web-based interactive tool is developed for offering prediction-level interpretation and gathering more details about specific nanopriming treatments. This work offers a promising framework for accelerating the agricultural applications of nanomaterials and may profoundly contribute to nanosafety assessment.

Zhiyu Zhang, Da Liu, Shengqiang Liu et al.

Contrastive learning has become one of the most impressive approaches for multi-modal representation learning. However, previous multi-modal works mainly focused on cross-modal understanding, ignoring in-modal contrastive learning, which limits the representation of each modality. In this paper, we propose a novel contrastive learning strategy, called $Turbo$, to promote multi-modal understanding by joint in-modal and cross-modal contrastive learning. Specifically, multi-modal data pairs are sent through the forward pass twice with different hidden dropout masks to get two different representations for each modality. With these representations, we obtain multiple in-modal and cross-modal contrastive objectives for training. Finally, we combine the self-supervised Turbo with the supervised multi-modal classification and demonstrate its effectiveness on two audio-text classification tasks, where the state-of-the-art performance is achieved on a speech emotion recognition benchmark dataset.