Papers/2609.17575
🧪 Test?View on arXiv

Temperon: Full-Time SAM Quality at a Third Less Wall-Clock

Not provided in the abstract

fine-tuningtraining optimizationsharpness-aware minimization
2609.17575
Builder Relevance
80%
1h ago

Abstract

Temperon achieves full-time sharpness-aware minimization (SAM) quality while reducing wall-clock time by a third through a novel training strategy.

Reality Card

Core Claim

Temperon matches the best full-time-SAM recipe on accuracy while achieving significant time savings, reaching target performance 35%, 34%, and 32% sooner on three datasets.

Method / Result

Temperon reaches target performance 35% sooner on CIFAR-10 compared to full-time SAM.

Limitations

The hand-off timing cannot be determined from the trajectory, making it less straightforward to reproduce the results.

Paper to code

Verified implementation resources so builders can test the paper’s claims instead of stopping at the abstract.

No verified implementation link has been attached yet. AIBuzzHub will keep this panel separate from unverified search results.
← Back to all papers