# Concertina: Data-Centric Adaptive Pipeline Parallelism for Efficient Heterogeneous Long-Context LLM Training

arXiv: https://arxiv.org/abs/2509.21275 · HTML: https://mlsanity.com/paper/2509.21275/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 21 of its 36 references that are in the map.
- First posted on arXiv 25 Sep 2025.

## Builds on (most cited first, 21 of 21)

- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [Llama 2: Open Foundation and Fine-Tuned Chat Models](https://mlsanity.com/paper/2307.09288/) (2023)
- [FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness](https://mlsanity.com/paper/2205.14135/) (2022)
- [FlashAttention-2: Faster Attention with Better Parallelism and Work Partitioning](https://mlsanity.com/paper/2307.08691/) (2023)
- [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://mlsanity.com/paper/1909.08053/) (2019)
- [DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model](https://mlsanity.com/paper/2405.04434/) (2024)
- [Ring Attention with Blockwise Transformers for Near-Infinite Context](https://mlsanity.com/paper/2310.01889/) (2023)
- [DeepSpeed Ulysses: System Optimizations for Enabling Training of Extreme Long Sequence Transformer Models](https://mlsanity.com/paper/2309.14509/) (2023)
- [MiniMax-01: Scaling Foundation Models with Lightning Attention](https://mlsanity.com/paper/2501.08313/) (2025)
- [DAPPLE: A Pipelined Data Parallel Approach for Training Large Models](https://mlsanity.com/paper/2007.01045/) (2020)
- [TeraPipe: Token-Level Pipeline Parallelism for Training Large-Scale Language Models](https://mlsanity.com/paper/2102.07988/) (2021)
- [Chimera: Efficiently Training Large-Scale Neural Networks with Bidirectional Pipelines](https://mlsanity.com/paper/2107.06925/) (2021)
- [Striped Attention: Faster Ring Attention for Causal Transformers](https://mlsanity.com/paper/2311.09431/) (2023)
- [FlexMoE: Scaling Large-scale Sparse Pre-trained Model Training via Dynamic Device Placement](https://mlsanity.com/paper/2304.03946/) (2023)
- [Optimal checkpointing for heterogeneous chains: how to train deep neural networks with limited memory](https://mlsanity.com/paper/1911.13214/) (2019)
- [Hanayo: Harnessing Wave-like Pipeline Parallelism for Enhanced Large Model Training Efficiency](https://mlsanity.com/paper/2308.15762/) (2023)
- [WLB-LLM: Workload-Balanced 4D Parallelism for Large Language Model Training](https://mlsanity.com/paper/2503.17924/) (2025)
- [DynaPipe: Optimizing Multi-task Training through Dynamic Pipelines](https://mlsanity.com/paper/2311.10418/) (2023)
- [Seq1F1B: Efficient Sequence-Level Pipeline Parallelism for Large Language Model Training](https://mlsanity.com/paper/2406.03488/) (2024)
- [GraphPipe: Improving Performance and Scalability of DNN Training with Graph Pipeline Parallelism](https://mlsanity.com/paper/2406.17145/) (2024)
- [SlimPipe: Memory-Thrifty and Efficient Pipeline Parallelism for Long-Context LLM Training](https://mlsanity.com/paper/2504.14519/) (2025)

## Abstract (arXiv)

Long context training is crucial for extending LLM context windows. Existing schemes, such as sequence parallelism, incur substantial communication overhead. Pipeline parallelism (PP) reduces this cost, but its effectiveness hinges on partitioning granularity. Batch-level PP employing sequence packing exhibits high memory consumption in long-context scenarios, whereas token-level PP splitting sequences into slices alleviates memory overhead but may introduce performance degradation. Moreover, the skewed sequence-length distribution in real-world datasets defeats any monolithic, static choice of PP granularity. In this paper, we propose textitDynamic Pipeline Parallelism (DPP), which transforms PP granularity from a static design choice into a workload-adaptive optimization space over packed, split, and hybrid chunks. DPP further introduces a new coupling between heterogeneous pipeline scheduling and gradient checkpointing. To solve this coupling, name co-optimizes dynamic chunk scheduling with textitStage-Aware Chunk-Level Adaptive Checkpointing. Comprehensive experiments demonstrate that name achieves up to 1.69texttimes speedup over FlexSP and up to 1.40texttimes over MEPipe. The source code is available at https://github.com/wsjdsg/InfiniPipe-code.