Our AlexNet model is a slight variation of the Caffe implementation of AlexNet (https://github.com/BVLC/caffe/tree/master/models/bvlc_alexnet). We removed model parallelism since nowadays most modern GPUs can hold the model in memory. Our model's accuracy is about 59.9% for top-1 category and 82.2% for top-5 categories, using just the center crop. In comparison, the BLVC AlexNet accuracy is 57.1% for top-1 category and 80.2% for top-5 categories. Assuming the ImageNet data folder has been correctly set up, you may achieve our accuracy numbers by launching the command:
python AlexNet_ImageNet_Distributed.py
You may use this python script to train AlexNet on multiple GPUs or machines. For a reference on distributed training, please check here. For example, the command for distributed training on the same machine (with multiple GPUs) with Windows is:
mpiexec -n <#workers> python AlexNet_ImageNet_Distributed.py