Graph-based Retrieval-Augmented Generation (GraphRAG) has emerged as a promising paradigm for enhancing LLM reliability by enabling multi-hop reasoning over graph-structured knowledge. However, existing LLMs struggle to efficiently process graph-structured inputs, as traditional attention mechanisms are sequence-based and introduce significant redundancy when serializing graphs into prompt sequences, leading to excessive computation and memory overhead. To address this, we introduce dependency attention, a novel graph-aware attention mechanism that restricts attention computation to token pairs with structural dependencies in the retrieved subgraph. Unlike standard self-attention that computes fully connected interactions, dependency attention prunes irrelevant token pairs and reuses computations along shared relational paths, substantially reducing inference overhead. Building on this idea, we develop DepCache, a KV cache management framework tailored for dependency attention. DepCache enables efficient KV cache reuse through (i) a graph-based KV cache reuse strategy that aligns KV caches across varying prompt contexts, enabling efficient cross-request reuse in GraphRAG, and (ii) a locality-aware replacement policy that leverages spatial and temporal access patterns to improve KV cache hit rate. Evaluations across diverse models and datasets show that DepCache improves LLM inference throughput by 1.5×-5.0× and reduces time-to-first-token latency by up to 3.2×, without compromising generation accuracy.

Graph neural networks (GNNs) have emerged as a promising method for learning from graph data, but large-scale GNN training requires extensive memory and computation resources. To address this, researchers have proposed using multi-GPU processing, which partitions graph data across GPUs for parallel training. However, vertex dependencies in multi-GPU GNN training lead to significant neighbor replications across GPUs, increasing memory consumption. The substantial intermediate data generated during training further exacerbates this issue. Neighbor replication and intermediate data constitute the primary memory consumption in GNN training (i.e., typically accounting for over 80%). In this work, we propose GNN task parallelism for multi-GPU GNN training, which reduces neighbor replication by partitioning training tasks in each layer across different GPUs rather than partitioning the graph structure. This approach only partitions the graph data within individual GPUs, reducing the memory requirements of single tasks while overlapping subgraph computation across different GPUs. Shared neighbor embeddings among different subgraphs can be efficiently reused within a single GPU. Additionally, we employ a task-decoupled GNN training framework, which decouples different training tasks to manage their associated intermediate data independently and release it as early as possible to reduce memory usage. By integrating these techniques, we propose a multi-GPU GNN training system, NeutronTask. Experimental results on a 4×A5000 GPU server show that NeutronTask effectively supports billion-scale full-graph GNN training. For small graphs where the training data fits into the GPUs, NeutronTask achieves 1.27× - 5.47× speedup compared to state-of-the-art GNN systems including NeutronStar and Sancus.

Graph neural networks (GNNs) have emerged as a promising direction. Training large-scale graphs that relies on distributed computing power poses new challenges. Existing distributed GNN systems leverage data parallelism by partitioning the input graph and distributing it to multiple workers. However, due to the irregular nature of the graph structure, existing distributed approaches suffer from unbalanced workloads and high overhead in managing cross-worker vertex dependencies. In this paper, we leverage tensor parallelism for distributed GNN training. GNN tensor parallelism eliminates cross-worker vertex dependencies by partitioning features instead of graph structures. Different workers are assigned training tasks on different feature slices with the same dimensional size, leading to a complete load balance. We achieve efficient GNN tensor parallelism through two critical functions. Firstly, we employ a generalized decoupled training framework to decouple NN operations from graph aggregation operations, significantly reducing the communication overhead caused by NN operations which must be computed using complete features. Secondly, we employ a memory-efficient task scheduling strategy to support the training of large graphs exceeding single GPU memory, while further improving performance by overlapping communication and computation. By integrating the above techniques, we propose a distributed GNN training system NeutronTP. Our experimental results on a 16-node Aliyun cluster demonstrate that NeutronTP achieves 1.29×-8.72× speedup over state-of-the-art GNN systems including DistDGL, NeutronStar, and Sancus.

Existing Graph Neural Network (GNN) training frameworks have been designed to help developers easily create performant GNN implementations. However, most existing GNN frameworks assume that the input graphs are static, but ignore that most real-world graphs are constantly evolving. Though many dynamic GNN models have emerged to learn from evolving graphs, the training process of these dynamic GNNs is dramatically different from traditional GNNs in that it captures both the spatial and temporal dependencies of graph updates. This poses new challenges for designing dynamic GNN training frameworks. First, the traditional batched training method fails to capture real-time structural evolution information. Second, the time-dependent nature makes parallel training hard to design. Third, it lacks system supports for users to efficiently implement dynamic GNNs. In this paper, we present NeutronStream, a framework for training dynamic GNN models. NeutronStream abstracts the input dynamic graph into a chronologically updated stream of events and processes the stream with an optimized sliding window to incrementally capture the spatial-temporal dependencies of events. Furthermore, NeutronStream provides a parallel execution engine to tackle the sequential event processing challenge to achieve high performance. NeutronStream also integrates a built-in graph storage structure that supports dynamic updates and provides a set of easy-to-use APIs that allow users to express their dynamic GNNs. Our experimental results demonstrate that, compared to state-of-the-art dynamic GNN implementations, NeutronStream achieves speedups ranging from 1.48X to 5.87X and an average accuracy improvement of 3.97%.