This repo contains the logit outputs produced by 61,565 independently and identically trained ResNets on the CIFAR-10 test-set.
To plot the histogram of accuracies across the first 500 trained models, run the following:
import numpy as np
import matplotlib.pyplot as plt
from huggingface_hub import HfApi
api = HfApi()
logits_path = api.hf_hub_download('kjj0/cifar10-multirun-logits-60k', repo_type='dataset', filename='logits500.npy')
labels_path =… See the full description on the dataset page:
https://huggingface.co/datasets/kjj0/cifar10-multirun-logits-60k.
