11 papers
PEARL: Adaptive Prefill-Decode Execution with Elasticity for Agentic Reinforcement Learning
Jiaan Zhu, Wei Gao, Youhui Bai +6
Multi-turn rollout dominates the cost of agentic reinforcement learning (RL). Asynchronous execution and elastic GPU resources can accelerate this stage, but adding rollout replica…
WaveAlign: Cache-Aware Query-Row Scheduling for Sparse Attention in Long-Video Generation
Zijian Dai, Sen Han, Youhui Bai +6
Long-video generation with diffusion transformers (DiTs) produces extremely long token sequences, making attention a dominant inference bottleneck. Dynamic sparse attention reduces…
DeaMoE: Efficient MoE Structure for Fast Small-Batch Decoding
Zewen Jin, Shen Fu, Zeping Duan +8
Mixture-of-Experts (MoE) models have been widely adopted in real-time interactive applications such as coding assistants, real-time audio-video interaction systems. To meet the ext…
HyperParallel-MoE: Multi-Core Interleaved Scheduling for Fast MoE Training on Ascend NPUs
Zewen Jin, Congkun Ai, Guangpeng Zhang +7
Modern Mixture-of-Experts (MoE) models increasingly rely on large-scale AI accelerator clusters for efficient training. Ascend NPUs expose heterogeneous on-chip compute resources,…
Accelerating Long-Tail Generation in Synchronous RLHF Training via Adaptive Tensor Parallelism
Long Zhao, Qinghe Wang, Jiaan Zhu +5
Reinforcement Learning from Human Feedback (RLHF) has become a key post-training paradigm for improving model quality. However, the synchronous three-stage RLHF pipeline is often b…
AdaCluster: Adaptive Query-Key Clustering for Sparse Attention in Video Generation
Haoyue Tan, Shengnan Wang, Yulin Qiao +5
Video diffusion transformers (DiTs) suffer from prohibitive inference latency due to quadratic attention complexity. Existing sparse attention methods either overlook semantic simi…
Lagom: Unleashing the Power of Communication and Computation Overlapping for Distributed LLM Training
Guanbin Xu, ZhenGuo Xu, Yuzhe Li +4
Overlapping communication with computation is crucial for distributed large-model training, yet optimizing it - especially when computation becomes the bottleneck-remains challengi…
LiteCache: A Query Similarity-Driven, GPU-Centric KVCache Subsystem for Efficient LLM Inference
Jiawei Yi, Ping Gong, Youhui Bai +10
During LLM inference, KVCache memory usage grows linearly with sequence length and batch size and often exceeds GPU capacity. Recent proposals offload KV states to host memory and…
HATA: Trainable and Hardware-Efficient Hash-Aware Top-k Attention for Scalable Large Model Inference
Ping Gong, Jiawei Yi, Shengnan Wang +13
Large Language Models (LLMs) have emerged as a pivotal research area, yet the attention module remains a critical bottleneck in LLM inference, even with techniques like KVCache to…
CentroidKV: Efficient Long-Context LLM Inference via KV Cache Clustering
Jie Hu, Shengnan Wang, Yutong He +8
Large language models (LLMs) with extended context windows have become increasingly prevalent for tackling complex tasks. However, the substantial Key-Value (KV) cache required for…
BigMac: A Communication-Efficient Mixture-of-Experts Model Structure for Fast Training and Inference
Zewen Jin, Shengnan Wang, Jiaan Zhu +5
The Mixture-of-Experts (MoE) structure scales the Transformer-based large language models (LLMs) and improves their performance with only the sub-linear increase in computation res…