DynaResize: Runtime GPU Reallocation for Disaggregated LLM Post-Training
Not on the map. Why?
Abstract (arXiv)
RL-based LLM post-training increasingly disaggregates Rollout and Training across separate GPU resources, but static GPU partitioning suffers from severe pipeline bubbles under long-tail rollout latency. We present DynaResize, a runtime GPU reallocation system that dynamically switches GPUs between Rollout and Training to balance stage execution times without changing RL semantics. DynaResize decomposes resizing into fine-grained operations and removes non-startup-critical work from the critical path through communicator reuse, bounded state staging, and hysteresis-based resizing. Experimental results show that DynaResize can improve end-to-end throughput by 66.5% and reduce total execution time by 33% over the optimal static configuration, while hiding 27% of role-switching overhead.
Key facts. Cited by 2 papers in the map (data through 14 Oct 2026). Builds on 29 of its 39 references that are in the map. First posted on arXiv 15 Jun 2026.
Builds on
29 of 39 references.
References · 29
- 2019PyTorch: An Imperative Style, High-Performance Deep Learning Librarycited by 20,305
- 2022Training language models to follow instructions with human feedbackcited by 15,971
- 2017Proximal Policy Optimization Algorithmscited by 15,215
- 2024The Llama 3 Herd of Modelscited by 14,715
- 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 8,065
- 2023Efficient Memory Management for Large Language Model Serving with PagedAttentioncited by 5,944
- 2025DAPO: An Open-Source LLM Reinforcement Learning System at Scalecited by 2,810
- 2024HybridFlow: A Flexible and Efficient RLHF Frameworkcited by 2,057
- 2019Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelismcited by 1,868
- 2019ZeRO: Memory Optimizations Toward Training Trillion Parameter Modelscited by 1,301
- 2026DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligencecited by 1,040
- 2023PyTorch FSDP: Experiences on Scaling Fully Sharded Data Parallelcited by 660
17 more
- 2023FlexGen: High-Throughput Generative Inference of Large Language Models with a Single GPUcited by 575
- 2017Ray: A Distributed Framework for Emerging AI Applicationscited by 527
- 2021ZeRO-Offload: Democratizing Billion-Scale Model Trainingcited by 341
- 2021ZeRO-Infinity: Breaking the GPU Memory Wall for Extreme Scale Deep Learningcited by 256
- 2025AReaL: A Large-Scale Asynchronous Reinforcement Learning System for Language Reasoningcited by 250
- 2025RLinf: Flexible and Efficient Large-scale Reinforcement Learning via Macro-to-Micro Flow Transformationcited by 57
- 2025Training-Free Group Relative Policy Optimizationcited by 55
- 2022Bamboo: Making Preemptible Instances Resilient for Affordable Training of Large DNNscited by 49
- 2025AsyncFlow: An Asynchronous Streaming RL Framework for Efficient LLM Post-Trainingcited by 41
- 2023Oobleck: Resilient Distributed Training of Large Models Using Pipeline Templatescited by 40
- 2024ReaL: Efficient RLHF Training of Large Language Models with Parameter Reallocationcited by 37
- 2025Laminar: A Scalable Asynchronous RL Post-Training Frameworkcited by 37
- 2025APRIL: Active Partial Rollouts in Reinforcement Learning to Tame Long-tail Generationcited by 30
- 2026ProRL Agent: Rollout-as-a-Service for RL Training of Multi-Turn LLM Agentscited by 20
- 2024Multi-Bin Batching for Increasing LLM Inference Throughputcited by 9
- 2026ROSE: Rollout On Serving GPUs via Cooperative Elasticity for Agentic RLcited by 8
- 2026SortedRL: Accelerating RL Training for LLMs through Online Length-Aware Schedulingcited by 6
Built on by
Citing papers per year: 2026: 2.
2 papers.