Convnet Performance Benchmark for PyTorch

usage

### the script runs on GPU when GPU is available, otherwise on CPU
### for training performance (batched mode)
./run.sh
### for inference performance (batched mode)
./run.sh --inference
### for inference latency (single batch)
./run.sh --inference --single

### for training performance (batched mode)
./run.sh --mkldnn
### for inference performance using mkldnn tensor (batched mode)
./run.sh --inference --mkldnn
### for inference latency using mkldnn tensor (single batch)
./run.sh --inference --single --mkldnn