edadaltocg / vit_base_patch16_224_in21k_ft_svhn

Model Card for Model ID

This model is a small timm/vit_base_patch16_224.orig_in21k_ft_in1k trained on svhn.

• Test Accuracy: 0.9672326367547633
• License: MIT

How to Get Started with the Model

Use the code below to get started with the model.

import timm
import torch
from torch import nn

model = timm.create_model("timm/vit_base_patch16_224.orig_in21k_ft_in1k",
pretrained=False)
model.head = nn.Linear(model.head.in_features, 10)
model.load_state_dict(
    torch.hub.load_state_dict_from_url(
        "https://huggingface.co/edadaltocg/vit_base_patch16_224_in21k_ft_svhn/resolve/main/pytorch_model.bin",
        map_location="cpu",
        file_name="vit_base_patch16_224_in21k_ft_svhn.pth",
    )
)

Training Data

Training data is svhn.

Training Hyperparameters

• config : scripts/train_configs/ft_svhn.json

• model : vit_base_patch16_224_in21k_ft_svhn

• dataset : svhn

• batch_size : 64

• epochs : 10

• validation_frequency : 1

• seed : 1

• criterion : CrossEntropyLoss

• criterion_kwargs : {}

• optimizer : SGD

• lr : 0.01

• optimizer_kwargs : {'momentum': 0.9, 'weight_decay': 0.0}

• scheduler : CosineAnnealingLR

• scheduler_kwargs : {'T_max': 10}

• debug : False

Testing Data

Testing data is svhn.

This model card was created by Eduardo Dadalto.