SkipPipe: Partial and Reordered Pipelining Framework for Training LLMs in Heterogeneous Networks

TL;DR

SkipPipe introduces a novel partial pipeline training framework for LLMs that reduces training time and enhances robustness by enabling stage skipping and reordering, with proven efficiency on large models.

Contribution

It is the first framework to enable stage skipping and reordering in pipeline training of LLMs, optimizing training speed while maintaining convergence.

Findings

Reduces training iteration time by up to 55%

Improves resistance to layer omission during inference

Effective on models from 500M to 8B parameters

Abstract

Data and pipeline parallelism are ubiquitous for training of Large Language Models (LLM) on distributed nodes. Driven by the need for cost-effective training, recent work explores efficient communication arrangement for end to end training. Motivated by LLM's resistance to layer skipping and layer reordering, in this paper, we explore stage (several consecutive layers) skipping in pipeline training, and challenge the conventional practice of sequential pipeline execution. We derive convergence and throughput constraints (guidelines) for pipelining with skipping and swapping pipeline stages. Based on these constraints, we propose SkipPipe, the first partial pipeline framework to reduce the end-to-end training time for LLMs while preserving the convergence. The core of SkipPipe is a path scheduling algorithm that optimizes the paths for individual microbatches and reduces idle time (due to microbatch collisions) on the distributed nodes, complying with the given stage skipping ratio. We extensively evaluate SkipPipe on LLaMa models from 500M to 8B parameters on up to 20 nodes. Our results show that SkipPipe reduces training iteration time by up to $55\%$ compared to full pipeline. Our partial pipeline training also improves resistance to layer omission during inference, experiencing a drop in perplexity of only $7\%$ when running only half the model. Our code is available at https://github.com/gensyn-ai/skippipe.