def _main():
base_dir = os.path.dirname(__file__)
request = {
"dataset":
"imagenet",
"model_name":
"resnet_50",
"data_dir":
os.path.join(base_dir, "./data/imagenet"),
"batch_size":
256,
"batch_size_val":
100,
"learning_rate":
0.1,
"epochs":
120,
"checkpoint_path":
os.path.join(base_dir, "./models_ckpt/resnet_50_imagenet_pruned"),
"checkpoint_save_period":
5, # save a checkpoint every 5 epoch
"checkpoint_eval_path":
os.path.join(base_dir, "./models_eval_ckpt/resnet_50_imagenet_pruned"),
"scheduler":
"uniform_auto",
"scheduler_file_name":
"resnet_50_imagenet_bn.yaml"
}
prune_model(request)
def _main():
base_dir = os.path.dirname(__file__)
request = {
"dataset":
"cifar10",
"model_name":
"vgg_m_16",
"data_dir":
os.path.join(base_dir, "./data/cifar10"),
"batch_size":
128,
"batch_size_val":
100,
"learning_rate":
0.1,
"epochs":
160,
"checkpoint_path":
os.path.join(base_dir, "./models_ckpt/vgg_m_16_cifar10"),
"checkpoint_save_period":
20, # save a checkpoint every epoch
"checkpoint_eval_path":
os.path.join(base_dir, "./models_eval_ckpt/vgg_m_16_cifar10"),
"scheduler":
"train"
}
prune_model(request)
def _main():
base_dir = os.path.dirname(__file__)
request = {
"dataset":
"imagenet",
"model_name":
"mobilenet_v2",
"data_dir":
os.path.join(base_dir, "./data/imagenet"),
"batch_size":
256,
"batch_size_val":
100,
"learning_rate":
0.05,
"epochs":
240,
"checkpoint_path":
os.path.join(base_dir, "./models_ckpt/mobilenet_v2_imagenet"),
"checkpoint_save_period":
5, # save a checkpoint every 5 epoch
"checkpoint_eval_path":
os.path.join(base_dir, "./models_eval_ckpt/mobilenet_v2_imagenet"),
"scheduler":
"train"
}
prune_model(request)
def _main():
base_dir = os.path.dirname(__file__)
request = {
"dataset":
"imagenet",
"model_name":
"resnet_50",
"data_dir":
os.path.join(base_dir, "/data/imagenet/tfrecord-dataset"),
"batch_size":
256,
"batch_size_val":
100,
"learning_rate":
0.1,
"epochs":
360,
"checkpoint_path":
os.path.join(base_dir, "./models_ckpt/resnet_50_imagenet_pruned"),
"checkpoint_save_period":
5, # save a checkpoint every 5 epoch
"checkpoint_eval_path":
os.path.join(base_dir, "./models_eval_ckpt/resnet_50_imagenet_pruned"),
"scheduler":
"uniform_auto",
"is_distill":
True,
"scheduler_file_name":
"resnet_50_imagenet_0.5_distill.yaml"
}
os.environ['L2_WEIGHT_DECAY'] = "5e-5"
prune_model(request)
def _main():
base_dir = os.path.dirname(__file__)
request = {
"dataset":
"mnist",
"model_name":
"lenet",
"data_dir":
os.path.join(base_dir, "./data/mnist"),
"batch_size":
120,
"batch_size_val":
100,
"learning_rate":
0.001,
"epochs":
12,
"checkpoint_path":
os.path.join(base_dir, "./models_ckpt/lenet_mnist_pruned"),
"checkpoint_save_period":
1, # save a checkpoint every n epoch
"checkpoint_eval_path":
os.path.join(base_dir, "./models_eval_ckpt/lenet_mnist_pruned"),
"scheduler":
"uniform_auto"
}
prune_model(request)