arXiv:2606. 30634v1 Announce Type: new Abstract: Modern large-scale LLM pretraining benefits from utilizing Pipeline Parallelism; however, synchronous implementations leave GPUs idle during pipeline bubbles, wasting computational resources.
Modern large-scale LLM pretraining benefits from utilizing Pipeline Parallelism; however, synchronous implementations leave GPUs idle during pipeline bubbles, wasting computational resources. Asynchronous Pipeline Parallelism eliminates these bubbles, maximizing throughput at the cost of gradient staleness.
arXiv:2606. 07881v1 Announce Type: new Abstract: Pipeline parallelism is essential for training large neural networks, but existing schedules trade off throughput, memory, and optimization consistency.
arXiv:2606. 03498v1 Announce Type: new Abstract: Training modern machine learning models increasingly requires computation to be distributed across many accelerators.
arXiv:2509.21275v5 Announce Type: replace-cross Abstract: Long context training is crucial for extending LLM context windows. Existing schemes, such as sequence parallelism, incur substantial communi...
arXiv:2509. 23722v2 Announce Type: replace-cross Abstract: Pipeline parallelism is widely used to train large language models (LLMs).
Para-Pipe is a hierarchical mapping framework that integrates intra- and inter-stage operator parallelism within a pipelined architecture for machine‑learning computational graphs on heterogeneous System‑on‑Chip (SoC) platforms. By selectively fine‑tuning parallelism levels across pipeline stages, it navigates the trade‑off between throughput and latency, reducing inter‑processor communication overhead and improving energy efficiency. Evaluation on Amlogic and Black Sesame SoCs shows multiple Pareto‑optimal configurations, with throughput‑optimized setups achieving up to 11.0% better energy efficiency than purely pipelined strategies and 23.3% better than non‑pipelined parallel execution.
The paper introduces a communication‑efficient method for adapting large language models on decentralized GPU meshes. It proposes an asynchronous two‑circuit system that uses fast compressed training with activation masking for pipeline‑parallel transfer and compressed data‑parallel synchronization, while a slower anchor circuit performs occasional unmasked passes. A spectral correction optimizer then denoises the masked gradients using these anchor priors, enabling high compression rates and achieving up to 40× throughput gains over internet‑grade connections while matching dense uncompressed performance.
The paper proposes a theoretical framework for scheduling high‑quality data in large language model training by extending functional scaling laws to account for time‑varying data quality. It identifies two regimes—noise‑limited and signal‑limited—where high‑quality data should be used differently, and introduces a Drop‑Stable‑Rampup training schedule that adjusts batch size at the quality transition. Experiments on 15B MoE and 600M dense models show significant accuracy gains over conventional decay schedules across multiple benchmarks.
Para‑Pipe is a hierarchical mapping framework that combines intra‑ and inter‑stage operator parallelism within a pipelined architecture to optimize deep‑learning inference on heterogeneous System‑on‑Chip (SoC) platforms. By selectively tuning parallelism levels across pipeline stages, it balances throughput and latency while reducing inter‑processor communication overhead. Evaluations on Amlogic and Black Sesame SoCs show Pareto‑optimal configurations, with throughput‑optimized settings achieving up to 11.0 % higher energy efficiency than purely pipelined approaches and 23.3 % over non‑pipelined parallel execution.
PipeLive introduces a method for live, in‑place reconfiguration of pipeline parallelism in large language model serving. By redesigning the KV cache layout and extending PageAttention, it enables dynamic resizing of the cache without interrupting inference. The system also uses an incremental KV patching mechanism to keep KV states consistent during reconfiguration, achieving significant reductions in reconfiguration time and improvements in latency metrics.
arXiv:2607. 24741v1 Announce Type: cross Abstract: Dynamic applications, including optimal-transport Flow Matching, repeatedly solve related entropic optimal transport problems, yet conventional distributed Sinkhorn processes frames sequentially and synchronizes after every iteration.
The paper presents a method for distributing large language model inference across multiple Intel AI PCs by splitting the model into pipeline shards, each pre‑compiled into an OpenVINO graph. Three key techniques—beam_idx Gather to enable GPU optimizations, speculative decoding on stateful models, and interleaved micro‑batching—allow a two‑node Llama 3.1 8B INT4 pipeline to serve two users at 1.79× the throughput of a single‑node model, while a four‑node deployment can run a 70B model that no single PC can hold. The authors provide code, benchmark logs, and reproduction scripts on GitHub.