Research

Paper

TESTING March 03, 2026

SENTINEL: Stagewise Integrity Verification for Pipeline Parallel Decentralized Training

Authors

Hadi Mohaghegh Dolatabadi, Thalaiyasingam Ajanthan, Sameera Ramasinghe, Chamin P Hewa Koneputugodage, Gil Avraham, Yan Zuo, Violetta Shevchenko, Alexander Long

Abstract

Decentralized training introduces critical security risks when executed across untrusted, geographically distributed nodes. While existing Byzantine-tolerant literature addresses data parallel (DP) training through robust aggregation methods, pipeline parallelism (PP) presents fundamentally distinct challenges. In PP, model layers are distributed across workers where the activations and their gradients flow between stages rather than being aggregated, making traditional DP approaches inapplicable. We propose SENTINEL, a verification mechanism for PP training without computation duplication. SENTINEL employs lightweight momentum-based monitoring using exponential moving averages (EMAs) to detect corrupted inter-stage communication. Unlike existing Byzantine-tolerant approaches for DP that aggregate parameter gradients across replicas, our approach verifies sequential activation/gradient transmission between layers. We provide theoretical convergence guarantees for this new setting that recovers classical convergence rates when relaxed to standard training. Experiments demonstrate successful training of up to 4B-parameter LLMs across untrusted distributed environments with up to 176 workers while maintaining model convergence and performance.

Metadata

arXiv ID: 2603.03592
Provider: ARXIV
Primary Category: cs.DC
Published: 2026-03-03
Fetched: 2026-03-05 06:06

Related papers

Raw Data (Debug)
{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2603.03592v1</id>\n    <title>SENTINEL: Stagewise Integrity Verification for Pipeline Parallel Decentralized Training</title>\n    <updated>2026-03-03T23:51:10Z</updated>\n    <link href='https://arxiv.org/abs/2603.03592v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2603.03592v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>Decentralized training introduces critical security risks when executed across untrusted, geographically distributed nodes. While existing Byzantine-tolerant literature addresses data parallel (DP) training through robust aggregation methods, pipeline parallelism (PP) presents fundamentally distinct challenges. In PP, model layers are distributed across workers where the activations and their gradients flow between stages rather than being aggregated, making traditional DP approaches inapplicable. We propose SENTINEL, a verification mechanism for PP training without computation duplication. SENTINEL employs lightweight momentum-based monitoring using exponential moving averages (EMAs) to detect corrupted inter-stage communication. Unlike existing Byzantine-tolerant approaches for DP that aggregate parameter gradients across replicas, our approach verifies sequential activation/gradient transmission between layers. We provide theoretical convergence guarantees for this new setting that recovers classical convergence rates when relaxed to standard training. Experiments demonstrate successful training of up to 4B-parameter LLMs across untrusted distributed environments with up to 176 workers while maintaining model convergence and performance.</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.DC'/>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.CR'/>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.LG'/>\n    <published>2026-03-03T23:51:10Z</published>\n    <arxiv:comment>70 pages, 22 figures, 20 tables</arxiv:comment>\n    <arxiv:primary_category term='cs.DC'/>\n    <author>\n      <name>Hadi Mohaghegh Dolatabadi</name>\n    </author>\n    <author>\n      <name>Thalaiyasingam Ajanthan</name>\n    </author>\n    <author>\n      <name>Sameera Ramasinghe</name>\n    </author>\n    <author>\n      <name>Chamin P Hewa Koneputugodage</name>\n    </author>\n    <author>\n      <name>Gil Avraham</name>\n    </author>\n    <author>\n      <name>Yan Zuo</name>\n    </author>\n    <author>\n      <name>Violetta Shevchenko</name>\n    </author>\n    <author>\n      <name>Alexander Long</name>\n    </author>\n  </entry>"
}