REVIEW 5 cited by
Unicron: Economizing Self-Healing LLM Training at Scale
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
Signed reviews
read the original abstract
Training large-scale language models is increasingly critical in various domains, but it is hindered by frequent failures, leading to significant time and economic costs. Current failure recovery methods in cloud-based settings inadequately address the diverse and complex scenarios that arise, focusing narrowly on erasing downtime for individual tasks without considering the overall cost impact on a cluster. We introduce Unicron, a workload manager designed for efficient self-healing in large-scale language model training. Unicron optimizes the training process by minimizing failure-related costs across multiple concurrent tasks within a cluster. Its key features include in-band error detection for real-time error identification without extra overhead, a dynamic cost-aware plan generation mechanism for optimal reconfiguration, and an efficient transition strategy to reduce downtime during state changes. Deployed on a 128-GPU distributed cluster, Unicron demonstrates up to a 1.9x improvement in training efficiency over state-of-the-art methods, significantly reducing failure recovery costs and enhancing the reliability of large-scale language model training.
Forward citations
Cited by 5 Pith papers
-
LLMPrism: Black-box Performance Diagnosis for Production LLM Training Platforms
A black-box system that reconstructs LLM training timelines and diagnoses slowdowns from switch-level network traffic flows alone.
-
Cloud abstractions for AI workloads
HarmonAIze proposes new cloud abstractions, split into micro- and macro-level control, to enable tenant-provider cooperation for more efficient and resilient AI workloads.
-
FlashRecovery: Fast and Low-Cost Recovery from Failures for Large-Scale Training of LLMs
FlashRecovery restores failed large-scale LLM training from data-parallel replicas in about 150 seconds on 4,800 devices, removing the need for periodic checkpointing.
-
Learning In Chaos: Efficient Autoscaling and Self-Healing for Multi-Party Distributed Training
A greedy shard-assignment protocol lets a joining node fetch model state from multiple neighbors in parallel, cutting scale-out delay to about one second and keeping other scaling events under 20 ms.
-
Digital Twin Networks for 6G Wireless Systems: Architecture, Enabling Technologies, Intelligent Control, and Open Challenges
Digital twin networks for 6G are classified into passive and active twins, with a comparison of computational complexity, use cases, and open challenges.
Discussion (0). Continue with ORCID to comment.