Training Trajectory Structure: Regime Detection, Cross-Seed Convergence, and Speculative Weight Prediction

Fuente: Zenodo
Gespeichert in:
Bibliographische Detailangaben
1. Verfasser: McEntire, Jeremy
Format: Recurso digital
Veröffentlicht: Zenodo 2026
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866901180297248768
author McEntire, Jeremy
author_facet McEntire, Jeremy
contents Training trajectories alternate between stable basins where gradients are predictable and chaotic transitions where they are not. We show these regimes are detectable via activation fingerprints, synchronized across independent seeds, and exploitable for speculative weight prediction. Consolidates and supersedes the Leap+Verify (DOI: 10.5281/zenodo.18828410) and Ensemble Collapse (DOI: 10.5281/zenodo.18828412) papers into a unified treatment. Linear prediction achieves 60-90% strict acceptance at 7B scale in stable regimes. Momentum-based prediction fails catastrophically at all scales (100-10,000x loss inflation). Phase boundaries synchronize across 5 independent seeds to within +/-50 steps. Final validation loss CV is 0.41% (124M), 2.43% (1.5B), 1.53% (7B).
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_19432379
institution Zenodo
language
publishDate 2026
publisher Zenodo
record_format zenodo
spellingShingle Training Trajectory Structure: Regime Detection, Cross-Seed Convergence, and Speculative Weight Prediction
McEntire, Jeremy
training trajectories
regime detection
speculative prediction
ensemble convergence
activation fingerprinting
Training trajectories alternate between stable basins where gradients are predictable and chaotic transitions where they are not. We show these regimes are detectable via activation fingerprints, synchronized across independent seeds, and exploitable for speculative weight prediction. Consolidates and supersedes the Leap+Verify (DOI: 10.5281/zenodo.18828410) and Ensemble Collapse (DOI: 10.5281/zenodo.18828412) papers into a unified treatment. Linear prediction achieves 60-90% strict acceptance at 7B scale in stable regimes. Momentum-based prediction fails catastrophically at all scales (100-10,000x loss inflation). Phase boundaries synchronize across 5 independent seeds to within +/-50 steps. Final validation loss CV is 0.41% (124M), 2.43% (1.5B), 1.53% (7B).
title Training Trajectory Structure: Regime Detection, Cross-Seed Convergence, and Speculative Weight Prediction
topic training trajectories
regime detection
speculative prediction
ensemble convergence
activation fingerprinting
url https://doi.org/10.5281/zenodo.19432379