LongVILA: Scaling Long-Context Visual Language Models for Long Videos

Long-context capability is critical for multi-modal foundation models, especially for long video understanding. We introduce LongVILA, a full-stack solution for long-context visual-language models by co-designing the algorithm and system. For model training, we upgrade existing VLMs to support long video understanding by incorporating two additional stages, i.e., long context extension and long supervised fine-tuning. However, training on long video is computationally and memory intensive. We introduce the long-context Multi-Modal Sequence Parallelism (MM-SP) system that efficiently parallelizes long video training and inference, enabling 2M context length training on 256 GPUs without any gradient checkpointing. LongVILA efficiently extends the number of video frames of VILA from 8 to 1024, improving the long video captioning score from 2.00 to 3.26 (out of 5), achieving 99.5% accuracy in 1400-frame (274k context length) video needle-in-a-haystack. LongVILA-8B demonstrates consistent accuracy improvements on long videos in the VideoMME benchmark as the number of frames increases. Besides, MM-SP is 2.1x - 5.7x faster than ring sequence parallelism and 1.1x - 1.4x faster than Megatron with context parallelism + tensor parallelism. Moreover, it seamlessly integrates with Hugging Face Transformers.

TorchGT: A Holistic System for Large-scale Graph Transformer Training

Graph Transformer is a new architecture that surpasses GNNs in graph learning. While there emerge inspiring algorithm advancements, their practical adoption is still limited, particularly on real-world graphs involving up to millions of nodes. We observe existing graph transformers fail on large-scale graphs mainly due to heavy computation, limited scalability and inferior model quality. Motivated by these observations, we propose TorchGT, the first efficient, scalable, and accurate graph transformer training system. TorchGT optimizes training at different levels. At algorithm level, by harnessing the graph sparsity, TorchGT introduces a Dual-interleaved Attention which is computation-efficient and accuracy-maintained. At runtime level, TorchGT scales training across workers with a communication-light Cluster-aware Graph Parallelism. At kernel level, an Elastic Computation Reformation further optimizes the computation by reducing memory access latency in a dynamic way. Extensive experiments demonstrate that TorchGT boosts training by up to 62.7x and supports graph sequence lengths of up to 1M.

Characterization of Large Language Model Development in the Datacenter

Large Language Models (LLMs) have presented impressive performance across several transformative tasks. However, it is non-trivial to efficiently utilize large-scale cluster resources to develop LLMs, often riddled with numerous challenges such as frequent hardware failures, intricate parallelization strategies, and imbalanced resource utilization. In this paper, we present an in-depth characterization study of a six-month LLM development workload trace collected from our GPU datacenter Acme. Specifically, we investigate discrepancies between LLMs and prior task-specific Deep Learning (DL) workloads, explore resource utilization patterns, and identify the impact of various job failures. Our analysis summarizes hurdles we encountered and uncovers potential opportunities to optimize systems tailored for LLMs. Furthermore, we introduce our system efforts: (1) fault-tolerant pretraining, which enhances fault tolerance through LLM-involved failure diagnosis and automatic recovery. (2) decoupled scheduling for evaluation, which achieves timely performance feedback via trial decomposition and scheduling optimization.

MEGA: A Memory-Efficient GNN Accelerator Exploiting Degree-Aware Mixed-Precision Quantization

Graph Neural Networks (GNNs) are becoming a promising technique in various domains due to their excellent capabilities in modeling non-Euclidean data. Although a spectrum of accelerators has been proposed to accelerate the inference of GNNs, our analysis demonstrates that the latency and energy consumption induced by DRAM access still significantly impedes the improvement of performance and energy efficiency. To address this issue, we propose a Memory-Efficient GNN Accelerator (MEGA) through algorithm and hardware co-design in this work. Specifically, at the algorithm level, through an in-depth analysis of the node property, we observe that the data-independent quantization in previous works is not optimal in terms of accuracy and memory efficiency. This motivates us to propose the Degree-Aware mixed-precision quantization method, in which a proper bitwidth is learned and allocated to a node according to its in-degree to compress GNNs as much as possible while maintaining accuracy. At the hardware level, we employ a heterogeneous architecture design in which the aggregation and combination phases are implemented separately with different dataflows. In order to boost the performance and energy efficiency, we also present an Adaptive-Package format to alleviate the storage overhead caused by the fine-grained bitwidth and diverse sparsity, and a Condense-Edge scheduling method to enhance the data locality and further alleviate the access irregularity induced by the extremely sparse adjacency matrix in the graph. We implement our MEGA accelerator in a 28nm technology node. Extensive experiments demonstrate that MEGA can achieve an average speedup of 38.3x, 7.1x, 4.0x, 3.6x and 47.6x, 7.2x, 5.4x, 4.5x energy savings over four state-of-the-art GNN accelerators, HyGCN, GCNAX, GROW, and SGCN, respectively, while retaining task accuracy.

Spiking NeRF: Making Bio-inspired Neural Networks See through the Real World

Spiking neuron networks (SNNs) have been thriving on numerous tasks to leverage their promising energy efficiency and exploit their potentialities as biologically plausible intelligence. Meanwhile, the Neural Radiance Fields (NeRF) render high-quality 3D scenes with massive energy consumption, and few works delve into the energy-saving solution with a bio-inspired approach. In this paper, we propose spiking NeRF (SpikingNeRF), which aligns the radiance ray with the temporal dimension of SNN, to naturally accommodate the SNN to the reconstruction of Radiance Fields. Thus, the computation turns into a spike-based, multiplication-free manner, reducing the energy consumption. In SpikingNeRF, each sampled point on the ray is matched onto a particular time step, and represented in a hybrid manner where the voxel grids are maintained as well. Based on the voxel grids, sampled points are determined whether to be masked for better training and inference. However, this operation also incurs irregular temporal length. We propose the temporal condensing-and-padding (TCP) strategy to tackle the masked samples to maintain regular temporal length, i.e., regular tensors, for hardware-friendly computation. Extensive experiments on a variety of datasets demonstrate that our method reduces the $76.74\%$ energy consumption on average and obtains comparable synthesis quality with the ANN baseline.

Boosting Distributed Full-graph GNN Training with Asynchronous One-bit Communication

Training Graph Neural Networks (GNNs) on large graphs is challenging due to the conflict between the high memory demand and limited GPU memory. Recently, distributed full-graph GNN training has been widely adopted to tackle this problem. However, the substantial inter-GPU communication overhead can cause severe throughput degradation. Existing communication compression techniques mainly focus on traditional DNN training, whose bottleneck lies in synchronizing gradients and parameters. We find they do not work well in distributed GNN training as the barrier is the layer-wise communication of features during the forward pass & feature gradients during the backward pass. To this end, we propose an efficient distributed GNN training framework Sylvie, which employs one-bit quantization technique in GNNs and further pipelines the curtailed communication with computation to enormously shrink the overhead while maintaining the model quality. In detail, Sylvie provides a lightweight Low-bit Module to quantize the sent data and dequantize the received data back to full precision values in each layer. Additionally, we propose a Bounded Staleness Adaptor to control the introduced staleness to achieve further performance enhancement. We conduct theoretical convergence analysis and extensive experiments on various models & datasets to demonstrate Sylvie can considerably boost the training throughput by up to 28.1x.

$\rm A^2Q$: Aggregation-Aware Quantization for Graph Neural Networks

As graph data size increases, the vast latency and memory consumption during inference pose a significant challenge to the real-world deployment of Graph Neural Networks (GNNs). While quantization is a powerful approach to reducing GNNs complexity, most previous works on GNNs quantization fail to exploit the unique characteristics of GNNs, suffering from severe accuracy degradation. Through an in-depth analysis of the topology of GNNs, we observe that the topology of the graph leads to significant differences between nodes, and most of the nodes in a graph appear to have a small aggregation value. Motivated by this, in this paper, we propose the Aggregation-Aware mixed-precision Quantization ($\rm A^2Q$) for GNNs, where an appropriate bitwidth is automatically learned and assigned to each node in the graph. To mitigate the vanishing gradient problem caused by sparse connections between nodes, we propose a Local Gradient method to serve the quantization error of the node features as the supervision during training. We also develop a Nearest Neighbor Strategy to deal with the generalization on unseen graphs. Extensive experiments on eight public node-level and graph-level datasets demonstrate the generality and robustness of our proposed method. Compared to the FP32 models, our method can achieve up to a 18.6x (i.e., 1.70bit) compression ratio with negligible accuracy degradation. Morever, compared to the state-of-the-art quantization method, our method can achieve up to 11.4\% and 9.5\% accuracy improvements on the node-level and graph-level tasks, respectively, and up to 2x speedup on a dedicated hardware accelerator.

DATE: Dual Assignment for End-to-End Fully Convolutional Object Detection

Fully convolutional detectors discard the one-to-many assignment and adopt a one-to-one assigning strategy to achieve end-to-end detection but suffer from the slow convergence issue. In this paper, we revisit these two assignment methods and find that bringing one-to-many assignment back to end-to-end fully convolutional detectors helps with model convergence. Based on this observation, we propose {\em \textbf{D}ual \textbf{A}ssignment} for end-to-end fully convolutional de\textbf{TE}ction (DATE). Our method constructs two branches with one-to-many and one-to-one assignment during training and speeds up the convergence of the one-to-one assignment branch by providing more supervision signals. DATE only uses the branch with the one-to-one matching strategy for model inference, which doesn't bring inference overhead. Experimental results show that Dual Assignment gives nontrivial improvements and speeds up model convergence upon OneNet and DeFCN. Code: https://github.com/YiqunChen1999/date.

PalQuant: Accelerating High-precision Networks on Low-precision Accelerators

Recently low-precision deep learning accelerators (DLAs) have become popular due to their advantages in chip area and energy consumption, yet the low-precision quantized models on these DLAs bring in severe accuracy degradation. One way to achieve both high accuracy and efficient inference is to deploy high-precision neural networks on low-precision DLAs, which is rarely studied. In this paper, we propose the PArallel Low-precision Quantization (PalQuant) method that approximates high-precision computations via learning parallel low-precision representations from scratch. In addition, we present a novel cyclic shuffle module to boost the cross-group information communication between parallel low-precision groups. Extensive experiments demonstrate that PalQuant has superior performance to state-of-the-art quantization methods in both accuracy and inference speed, e.g., for ResNet-18 network quantization, PalQuant can obtain 0.52\% higher accuracy and 1.78$\times$ speedup simultaneously over their 4-bit counter-part on a state-of-the-art 2-bit accelerator. Code is available at \url{https://github.com/huqinghao/PalQuant}.

Deep Learning Workload Scheduling in GPU Datacenters: Taxonomy, Challenges and Vision

Deep learning (DL) shows its prosperity in a wide variety of fields. The development of a DL model is a time-consuming and resource-intensive procedure. Hence, dedicated GPU accelerators have been collectively constructed into a GPU datacenter. An efficient scheduler design for such GPU datacenter is crucially important to reduce the operational cost and improve resource utilization. However, traditional approaches designed for big data or high performance computing workloads can not support DL workloads to fully utilize the GPU resources. Recently, substantial schedulers are proposed to tailor for DL workloads in GPU datacenters. This paper surveys existing research efforts for both training and inference workloads. We primarily present how existing schedulers facilitate the respective workloads from the scheduling objectives and resource consumption features. Finally, we prospect several promising future research directions. More detailed summary with the surveyed paper and code links can be found at our project website: https://github.com/S-Lab-System-Group/Awesome-DL-Scheduling-Papers