29 citations · 75 across the 13 of their papers we have counts for
15 papers
ReFlow: Replay-Driven Decoupling for Efficient RL Post-Training
Yukun Tian, Junwei Su, Haofeng Xu +6
Reinforcement learning (RL) has been a critical paradigm for improving large language model (LLM) reasoning and agentic capabilities, but its efficiency remains a major bottleneck.…
Reward-Aligned Reweighting for On-Policy Distillation
Haofeng Xu, Junwei Su, Lansong Diao +2
On-policy distillation (OPD) trains a student language model with dense feedback from a stronger teacher on student-generated trajectories. Yet standard OPD weights token-level dis…
Token Economics for LLM Agents: A Dual-View Study from Computing and Economics
Yuxi Chen, Junming Chen, Chenyu He +9
As LLM agents evolve, tokens have emerged as the core economic primitives of Agentic AI. However, their exponential consumption introduces severe computational, collaborative, and…
GAC: Stabilizing Asynchronous RL Training for LLMs via Gradient Alignment Control
Haofeng Xu, Junwei Su, Yukun Tian +3
Asynchronous execution is essential for scaling reinforcement learning (RL) to modern large model workloads, including large language models and AI agents, but it can fundamentally…
Adaptive Hybrid Caching for Efficient Text-to-Video Diffusion Model Acceleration
Yuanxin Wei, Lansong Diao, Bujiao Chen +6
Efficient video generation models are increasingly vital for multimedia synthetic content generation. Leveraging the Transformer architecture and the diffusion process, video DiT m…
SRDiffusion: Accelerate Video Diffusion Inference via Sketching-Rendering Cooperation
Shenggan Cheng, Yuanxin Wei, Lansong Diao +8
Leveraging the diffusion transformer (DiT) architecture, models like Sora, CogVideoX and Wan have achieved remarkable progress in text-to-video, image-to-video, and video editing t…
HAP: SPMD DNN Training on Heterogeneous GPU Clusters with Automated Program Synthesis
Shiwei Zhang, Lansong Diao, Chuan Wu +3
Single-Program-Multiple-Data (SPMD) parallelism has recently been adopted to train large deep neural networks (DNNs). Few studies have explored its applicability on heterogeneous c…
Ada-Grouper: Accelerating Pipeline Parallelism in Preempted Network by Adaptive Group-Scheduling for Micro-Batches
Siyu Wang, Zongyan Cao, Chang Si +3
Pipeline parallelism has been demonstrated to be a remarkable approach to improve throughput for training deep neural networks with billions of parameters over heterogeneous cluste…
Auto-Parallelizing Large Models with Rhino: A Systematic Approach on Production AI Platform
Shiwei Zhang, Lansong Diao, Siyu Wang +7
We present Rhino, a system for accelerating tensor programs with automatic parallelization on AI platform for real production environment. It transforms a tensor program written fo…
Expediting Distributed DNN Training with Device Topology-Aware Graph Deployment
Shiwei Zhang, Xiaodong Yi, Lansong Diao +3
This paper presents TAG, an automatic system to derive optimized DNN training graph and its deployment onto any device topology, for expedited training in device- and topology- het…
Optimizing DNN Compilation for Distributed Training with Joint OP and Tensor Fusion
Xiaodong Yi, Shiwei Zhang, Lansong Diao +6
This paper proposes DisCo, an automatic deep learning compilation module for data-parallel distributed training. Unlike most deep learning compilers that focus on training or infer…
DISC: A Dynamic Shape Compiler for Machine Learning Workloads
Kai Zhu, Wenyi Zhao, Zhen Zheng +8
Many recent machine learning models show dynamic shape characteristics. However, existing AI compiler optimization systems suffer a lot from problems brought by dynamic shape model…
FusionStitching: Boosting Memory Intensive Computations for Deep Learning Workloads
Zhen Zheng, Pengzhan Zhao, Guoping Long +6
We show in this work that memory intensive computations can result in severe performance problems due to off-chip memory access and CPU-GPU context switch overheads in a wide range…
Auto-MAP: A DQN Framework for Exploring Distributed Execution Plans for DNN Workloads
Siyu Wang, Yi Rong, Shiqing Fan +6
The last decade has witnessed growth in the computational requirements for training deep neural networks. Current approaches (e.g., data/model parallelism, pipeline parallelism) pa…
DAPPLE: A Pipelined Data Parallel Approach for Training Large Models
Shiqing Fan, Yi Rong, Chen Meng +10
It is a challenging task to train large DNN models on sophisticated GPU platforms with diversified interconnect capabilities. Recently, pipelined training has been proposed as an e…