benchmarks

Computational/deep learning benchmarks.

Currently, there is the only script called super_burner.py. It supports matmul, conv2d and rnn tasks in both PyTorch and TensorFlow.

Example usage:

CUDA_VISIBLE_DEVICES=0 python super_burner.py \
    --framework=tf --task=conv2d \
    --batch-size=32 --iterations=100000 \
    --kernel-size=3 --input-size-2d=224 --depth=64

To get information about other supported options, type python super_burner.py --help.