Model card for resnet18_cifar10

This is a resnet18 model trained on the cifar10 dataset. To load this model use the timm library and run the following code:

import timm
model = timm.create_model("hf_hub:SamAdamDay/resnet18_cifar10", pretrained=True)

The model was trained using the following command:

./distributed_train.sh --dataset torch/cifar10 --data-dir /root/data --dataset-download --model resnet18 --lr-base 0.3 --epochs 100 --input-size 3 256 256 -mean 0.49139968 0.48215827 0.44653124 --std 0.24703233 0.24348505 0.26158768 --num-classes 10

Metrics

The model has a test accuracy of 94.73.

Model Details

• Dataset: cifar10
• Number of epochs: 100
• Batch size: 128
• Base LR: 0.3
• LR scheduler: cosine
• Input size (3, 256, 256), images are scaled to this size
• PyTorch version: 2.3.0+cu121
• timm version: 1.0.7