Piper: A Programmable Distributed Training System

📅 2026-06-09
📈 Citations: 0
Influential: 0
📄 PDF
🤖 AI Summary
Existing large-scale model training systems struggle to flexibly compose diverse parallelization strategies, often relying on manual expert tuning and lacking generality. This work proposes a programmable distributed training system that enables users to declaratively specify composite parallelism strategies—such as data, pipeline, and expert parallelism—through model annotations and scheduling directives. These specifications are compiled via a unified intermediate representation (IR) into device-level execution plans, fully decoupling strategy definition from runtime execution over a global compute-communication DAG. The system is the first to support automatic compilation of user-defined composite strategies, matching the performance of established approaches like ZeRO while significantly improving both performance and memory efficiency in complex scenarios such as DeepSeek-V3’s DualPipe.
📝 Abstract
Large-scale model training increasingly relies on composing multiple parallelism strategies, such as data, pipeline, and expert parallelism, together with memory-saving optimizations like ZeRO. Deployed systems for foundation model pretraining often rely on human experts to manually design a high-level parallelism strategy then implement the corresponding low-level execution strategy, making it difficult to adapt the system to new strategies. Meanwhile, many general-purpose frameworks are more flexible but their implementations are still tied to a fixed set of common parallelism strategies, making it challenging to integrate state-of-the-art strategies. We present Piper, a user-controllable distributed training system that decouples the strategy from the runtime implementation. Piper allows users to declare a comprehensive distributed training strategy with a small set of model annotations and scheduling directives. Each directive applies a transformation on Piper's intermediate representation (IR), a unified global training DAG that represents all computation and communication. Using this IR, Piper compiles per-device execution plans and executes them with a distributed runtime agnostic to the strategy. We show that the combined system maintains performance parity on commonly available strategies such as ZeRO, while also enabling additional performance and memory efficiency gains through joint scheduling of compute and communication in composed parallelism strategies such as DeepSeek-V3's DualPipe.
Problem

Research questions and friction points this paper is trying to address.

distributed training
parallelism strategies
model parallelism
training system
flexibility
Innovation

Methods, ideas, or system contributions that make the work stand out.

programmable distributed training
strategy-runtime decoupling
intermediate representation (IR)
composed parallelism
joint compute-communication scheduling
🔎 Similar Papers
2024-06-07International Symposium on High-Performance Computer ArchitectureCitations: 5
M
Megan Frisella
University of Washington, Seattle, WA, USA
S
Shubham Tiwari
University of Washington, Seattle, WA, USA
A
Andy Ruan
University of Washington, Seattle, WA, USA
Y
Yi Pan
University of Washington and Shanghai Jiao Tong University, Seattle, WA, USA
P
Parker Gustafson
University of Washington, Seattle, WA, USA
M
Mat Jacob
University of Washington, Seattle, WA, USA
G
Gilbert Bernstein
University of Washington, Seattle, WA, USA
S
Stephanie Wang
University of Washington, Seattle, WA, USA