Anonymous ViT-B/32 CIFAR-10 Review Artifacts

This private repository contains the model artifacts supplied for anonymous peer review.

Contents

  • weights/model_0.pt through weights/model_71.pt: 72 individual ViT-B/32 checkpoints.
  • metadata/hparam_info.json: per-model training hyperparameters.
  • metadata/individual_model_result.json: per-model train, calibration, and test accuracy.
  • training_logs/model_0.txt through training_logs/model_71.txt: sanitized training logs.
  • release_manifest.json: machine-readable file inventory.

model_0 is the LP-initialization reference checkpoint. The other models use the hyperparameters recorded in metadata/hparam_info.json.

Dataset

The checkpoints are for CIFAR-10, with 10 output classes.

Anonymity

Author names, affiliations, local paths, hostnames, email addresses, and external experiment-tracking links have been removed from the released text metadata. Original local files were not modified.

Scope

This review release intentionally excludes ensemble-derived artifacts, including greedy_soup.pt.

Loading

The exact model-construction code and evaluation entry point should be provided in the anonymous code repository associated with the submission. Load these files only in an environment where PyTorch checkpoint files are trusted.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support