torchbenchmark/models/vgg16/__init__.py

from torchbenchmark.util.framework.vision.model_factory import TorchVisionModel
from torchbenchmark.tasks import COMPUTER_VISION

class Model(TorchVisionModel):
    task = COMPUTER_VISION.CLASSIFICATION

    # Original train batch size 256 on 4-GPU system
    # Downscale to 64 to run on single GPU device
    # Source: https://arxiv.org/pdf/1409.1556.pdf
    DEFAULT_TRAIN_BSIZE = 64
    DEFAULT_EVAL_BSIZE = 4

    def __init__(self, test, device, jit=False, batch_size=None, extra_args=[]):
        super().__init__(model_name="vgg16", test=test, device=device, jit=jit,
                         batch_size=batch_size, extra_args=extra_args)