Cover Image

Unleashing Efficient Asynchronous RL Post-Training via Staleness-Constrained Rollout Coordination

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Li, Haoyang, Lin, Sheng, Fu, Fangcheng, Zhou, Yuming, Ji, Xiaodong, Zhao, Yanfeng, Wang, Lefeng, Jiang, Jie, Cui, Bin
Format:	Preprint
Published:	2026
Subjects:	Distributed, Parallel, and Cluster Computing
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

EchoPFL: Asynchronous Personalized Federated Learning on Mobile Devices with On-Demand Staleness Control
by: Li, Xiaochen, et al.
Published: (2024)

Malleus: Straggler-Resilient Hybrid Parallel Training of Large-scale Models via Malleable Data and Model Parallelization
by: Li, Haoyang, et al.
Published: (2024)

LobRA: Multi-tenant Fine-tuning over Heterogeneous Data
by: Lin, Sheng, et al.
Published: (2025)

Hetu v2: A General and Scalable Deep Learning System with Hierarchical and Heterogeneous Single Program Multiple Data Annotations
by: Li, Haoyang, et al.
Published: (2025)

HexiScale: Facilitating Large Language Model Training over Heterogeneous Hardware
by: Yan, Ran, et al.
Published: (2024)

RollPacker: Mitigating Long-Tail Rollouts for Fast, Synchronous RL Post-Training
by: Gao, Wei, et al.
Published: (2025)

MSPipe: Efficient Temporal GNN Training via Staleness-Aware Pipeline
by: Sheng, Guangming, et al.
Published: (2024)

FedStaleWeight: Buffered Asynchronous Federated Learning with Fair Aggregation via Staleness Reweighting
by: Ma, Jeffrey, et al.
Published: (2024)

Efficient Multi-round LLM Inference over Disaggregated Serving
by: He, Wenhao, et al.
Published: (2026)

Hydraulis: Balancing Large Transformer Model Training via Co-designing Parallel Strategies and Data Assignment
by: Li, Haoyang, et al.
Published: (2024)

Memory Efficient and Staleness Free Pipeline Parallel DNN Training Framework with Improved Convergence Speed
by: Dutta, Ankita, et al.
Published: (2025)

Laminar: A Scalable Asynchronous RL Post-Training Framework
by: Sheng, Guangming, et al.
Published: (2025)

BOute: Cost-Efficient LLM Serving with Heterogeneous LLMs and GPUs via Multi-Objective Bayesian Optimization
by: Jiang, Youhe, et al.
Published: (2026)

TiMePReSt: Time and Memory Efficient Pipeline Parallel DNN Training with Removed Staleness
by: Dutta, Ankita, et al.
Published: (2024)

ROSE: Rollout On Serving GPUs via Cooperative Elasticity for Agentic RL
by: Gao, Wei, et al.
Published: (2026)

Spindle: Efficient Distributed Training of Multi-Task Large Models via Wavefront Scheduling
by: Wang, Yujie, et al.
Published: (2024)

LAER-MoE: Load-Adaptive Expert Re-layout for Efficient Mixture-of-Experts Training
by: Liu, Xinyi, et al.
Published: (2026)

LSH-MoE: Communication-efficient MoE Training via Locality-Sensitive Hashing
by: Nie, Xiaonan, et al.
Published: (2024)

DistFlow: A Fully Distributed RL Framework for Scalable and Efficient LLM Post-Training
by: Wang, Zhixin, et al.
Published: (2025)

InfiniPipe: Elastic Pipeline Parallelism for Efficient Variable-Length Long-Context LLM Training
by: Wang, Shiju, et al.
Published: (2025)

ThunderServe: High-performance and Cost-efficient LLM Serving in Cloud Environments
by: Jiang, Youhe, et al.
Published: (2025)

TridentServe: A Stage-level Serving System for Diffusion Pipelines
by: Xia, Yifei, et al.
Published: (2025)

AReaL-Hex: Accommodating Asynchronous RL Training over Heterogeneous GPUs
by: Yan, Ran, et al.
Published: (2025)

RollMux: Phase-Level Multiplexing for Disaggregated RL Post-Training
by: Wu, Tianyuan, et al.
Published: (2025)

A-3PO: Accelerating Asynchronous LLM Training with Staleness-aware Proximal Policy Approximation
by: Li, Xiaocan, et al.
Published: (2025)

Galvatron: An Automatic Distributed System for Efficient Foundation Model Training
by: Liu, Xinyi, et al.
Published: (2025)

Mitigating Staleness in Asynchronous Pipeline Parallelism via Basis Rotation
by: Jung, Hyunji, et al.
Published: (2026)

Staleness-Centric Optimizations for Parallel Diffusion MoE Inference
by: Luo, Jiajun, et al.
Published: (2024)

ABS: Adaptive Bounded Staleness Converges Faster and Communicates Less
by: Tan, Qiao, et al.
Published: (2023)

Boosting Blockchain Throughput: Parallel EVM Execution with Asynchronous Storage for Reddio
by: Qi, Xiaodong, et al.
Published: (2025)

Demystifying Cost-Efficiency in LLM Serving over Heterogeneous GPUs
by: Jiang, Youhe, et al.
Published: (2025)

Unleashing Multicore Strength for Efficient Execution of Transactions
by: Ravish, Ankit, et al.
Published: (2024)

Schedule-Level Shared-Prefix Reuse for LLM RL Training
by: Li, Pengbo, et al.
Published: (2026)

Cascadia: An Efficient Cascade Serving System for Large Language Models
by: Jiang, Youhe, et al.
Published: (2025)

KUBEDIRECT: Unleashing the Full Power of the Cluster Manager for Serverless Computing
by: Qi, Sheng, et al.
Published: (2026)

Lagom: Unleashing the Power of Communication and Computation Overlapping for Distributed LLM Training
by: Xu, Guanbin, et al.
Published: (2026)

OServe: Accelerating LLM Serving via Spatial-Temporal Workload Orchestration
by: Jiang, Youhe, et al.
Published: (2026)

APEX: Asynchronous Parallel CPU-GPU Execution for Online LLM Inference on Constrained GPUs
by: Fan, Jiakun, et al.
Published: (2025)

Improving Automatic Parallel Training via Balanced Memory Workload Optimization
by: Wang, Yujie, et al.
Published: (2023)

Unleashing Scalable Context Parallelism for Foundation Models Pre-Training via FCP
by: Zhao, Yilong, et al.
Published: (2026)