# MindSpeed RL: Distributed Dataflow for Scalable and Efficient RL Training on Ascend NPU Cluster

arXiv: https://arxiv.org/abs/2507.19017 · HTML: https://mlsanity.com/paper/2507.19017/

## Key facts

- Cited by 9 papers in the map (data through 14 Oct 2026).
- Placed in MoE in reinforcement learning and robotics, one of 12 directions in Mixture-of-experts.
- Builds on 25 of its 39 references that are in the map.
- First posted on arXiv 25 Jul 2025.

## Placed in

- Mixture-of-experts / [MoE in reinforcement learning and robotics](https://mlsanity.com/moe/moe-in-reinforcement-learning-and-robotics/)

## Builds on (most cited first, 25 of 25)

- [Measuring Mathematical Problem Solving With the MATH Dataset](https://mlsanity.com/paper/2103.03874/) (2021)
- [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://mlsanity.com/paper/2104.09864/) (2021)
- [DAPO: An Open-Source LLM Reinforcement Learning System at Scale](https://mlsanity.com/paper/2503.14476/) (2025)
- [HybridFlow: A Flexible and Efficient RLHF Framework](https://mlsanity.com/paper/2409.19256/) (2024)
- [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://mlsanity.com/paper/1909.08053/) (2019)
- [ZeRO: Memory Optimizations Toward Training Trillion Parameter Models](https://mlsanity.com/paper/1910.02054/) (2019)
- [Kimi k1.5: Scaling Reinforcement Learning with LLMs](https://mlsanity.com/paper/2501.12599/) (2025)
- [SGLang: Efficient Execution of Structured Language Model Programs](https://mlsanity.com/paper/2312.07104/) (2023)
- [PyTorch FSDP: Experiences on Scaling Fully Sharded Data Parallel](https://mlsanity.com/paper/2304.11277/) (2023)
- [Efficient Large-Scale Language Model Training on GPU Clusters Using Megatron-LM](https://mlsanity.com/paper/2104.04473/) (2021)
- [Ray: A Distributed Framework for Emerging AI Applications](https://mlsanity.com/paper/1712.05889/) (2017)
- [Ring Attention with Blockwise Transformers for Near-Infinite Context](https://mlsanity.com/paper/2310.01889/) (2023)
- [DeepSpeed-MoE: Advancing Mixture-of-Experts Inference and Training to Power Next-Generation AI Scale](https://mlsanity.com/paper/2201.05596/) (2022)
- [AReaL: A Large-Scale Asynchronous Reinforcement Learning System for Language Reasoning](https://mlsanity.com/paper/2505.24298/) (2025)
- [DeepSpeed-Chat: Easy, Fast and Affordable RLHF Training of ChatGPT-like Models at All Scales](https://mlsanity.com/paper/2308.01320/) (2023)
- [Colossal-AI: A Unified Deep Learning System For Large-Scale Parallel Training](https://mlsanity.com/paper/2110.14883/) (2021)
- [Reinforcement Learning Enhanced LLMs: A Survey](https://mlsanity.com/paper/2412.10400/) (2024)
- [Asynchronous RLHF: Faster and More Efficient Off-Policy RL for Language Models](https://mlsanity.com/paper/2410.18252/) (2024)
- [StreamRL: Scalable, Heterogeneous, and Elastic RL for LLMs with Disaggregated Stream Generation](https://mlsanity.com/paper/2504.15930/) (2025)
- [NeMo-Aligner: Scalable Toolkit for Efficient Model Alignment](https://mlsanity.com/paper/2405.01481/) (2024)
- [ReaL: Efficient RLHF Training of Large Language Models with Parameter Reallocation](https://mlsanity.com/paper/2406.14088/) (2024)
- [Exploring Data Scaling Trends and Effects in Reinforcement Learning from Human Feedback](https://mlsanity.com/paper/2503.22230/) (2025)
- [Jenga: Effective Memory Management for Serving LLM with Heterogeneity](https://mlsanity.com/paper/2503.18292/) (2025)
- [RLlib Flow: Distributed Reinforcement Learning is a Dataflow Problem](https://mlsanity.com/paper/2011.12719/) (2020)
- [Model Parallelism With Subnetwork Data Parallelism](https://mlsanity.com/paper/2507.09029/) (2025)

## Built on by (most cited first, 9 of 9)

- [A-3PO: Accelerating Asynchronous LLM Training with Staleness-aware Proximal Policy Approximation](https://mlsanity.com/paper/2512.06547/) (2025)
- [TideRL: Boosting Agentic RL Goodput with Readiness-Aware Scheduling](https://mlsanity.com/paper/2608.10402/) (2026)
- [MindVL: Towards Efficient and Effective Training of Multimodal Large Language Models on Ascend NPUs](https://mlsanity.com/paper/2509.11662/) (2025)
- [EDCO: Dynamic Curriculum Orchestration for Domain-specific Large Language Model Fine-tuning](https://mlsanity.com/paper/2601.03725/) (2026)
- [Agentic-KGR: Co-evolutionary Knowledge Graph Construction through Multi-Agent Reinforcement Learning](https://mlsanity.com/paper/2510.09156/) (2025)
- [Periodic Asynchrony: An On-Policy Approach for Accelerating LLM Reinforcement Learning](https://mlsanity.com/paper/2511.18871/) (2025)
- [Reinforcement Learning with Promising Tokens for Large Language Models](https://mlsanity.com/paper/2602.03195/) (2026)
- [Schedule-and-Calibrate: Utility-Guided Multi-Task Reinforcement Learning for Code LLMs](https://mlsanity.com/paper/2605.06111/) (2026)
- [Performance Foundations of Parallel & Distributed Reasoning Language Models](https://mlsanity.com/paper/2608.27046/) (2026)

## Abstract (arXiv)

Reinforcement learning (RL) is a paradigm increasingly used to align large language models. Popular RL algorithms utilize multiple workers and can be modeled as a graph, where each node is the status of a worker and each edge represents dataflow between nodes. Owing to the heavy cross-node dependencies, the RL training system usually suffers from poor cluster scalability and low memory utilization. In this article, we introduce MindSpeed RL, an effective and efficient system for large-scale RL training. Unlike existing centralized methods, MindSpeed RL organizes the essential data dependencies in RL training, i.e., sample flow and resharding flow, from a distributed view. On the one hand, a distributed transfer dock strategy, which sets controllers and warehouses on the basis of the conventional replay buffer, is designed to release the dispatch overhead in the sample flow. A practical allgather--swap strategy is presented to eliminate redundant memory usage in resharding flow. In addition, MindSpeed RL further integrates numerous parallelization strategies and acceleration techniques for systematic optimization. Compared with existing state-of-the-art systems, comprehensive experiments on the RL training of popular Qwen2.5-Dense-7B/32B, Qwen3-MoE-30B, and DeepSeek-R1-MoE-671B show that MindSpeed RL increases the throughput by 1.42 ~ 3.97 times. Finally, we open--source MindSpeed RL and perform all the experiments on a super pod of Ascend with 384 neural processing units (NPUs) to demonstrate the powerful performance and reliability of Ascend.