Preprint / scholarly article

Tail-Limited Useful Compute for Large-Scale AI Training

K. Takahashi

Published
DOI
10.5281/zenodo.18051330

Full text PDF (Zenodo)

Abstract

The preprint defines Tail-Limited Useful Compute, a contract-based telemetry framework that converts tail latency into fail-closed bottleneck certificates and anytime-valid lower bounds on useful progress for large-scale AI training.

Keywords

  • tail latency
  • useful compute
  • AI training
  • telemetry contract
  • fail-closed verification
  • confidence sequences
  • bottleneck certificates
  • stragglers
  • distributed training
  • auditability
  • throughput floors

Identifiers and source records