This project contains model weights for several deep neural networks trained on the Ecoset dataset with the PyTorch and Horovod frameworks. All models were trained using 20 A100 GPUs. We used a batch size of 256, momentum of 0.9, weight decay of 1e-4 and a learning rate of 0.001. Note, that the state_dict.pt weights were trained with a normalized cross entropy loss to counteract the inbalance. Weights that were trained with a standard cross entropy loss can also be found in the model folders. **VGG16** The VGG-16 model was trained without batch normalization. The model achieved a Top1 accuracy of 67.8% on the Ecoset test spit. You can find an example on how to run it in vgg16/predict.py. **Resnet50** The model was trained for 121 epochs. The model achieved a Top1 accuracy of 66.7% on the Ecoset test spit. You can find an example on how to run it in resnet50/predict.py. **Alexnet** The model was trained for 71 epochs. The model achieved a Top1 accuracy of 55.2% on the validation spit. You can find an example on how to run it in alexnet/predict.py. **Inception** The model was trained for 118 epochs. The model achieved a Top1 accuracy of 57.8% on the validation spit. You can find an example on how to run it in inception/predict.py.