This repo contains benchmarking scripts for training and inference on common CNN topologies in TensorFlow:
- Inception V3
- ResNet 50
- ResNet 152
- VGG 16
To initiate benchmarking, cd into this directory and run
bash intel_tf_cnn_benchmarks.sh for training or
bash intel_tf_cnn_benchmarks_inference.sh for inference. This will execute the Inception V3 model with BS=128 for training or BS=1 for inference. To run all topologies, pass
all as an argument to the script:
bash intel_tf_cnn_benchmarks.sh all
Each script takes the following actions:
- Detects which version of TensorFlow is present and clones the official TensorFlow benchmarks repo.
tf_cnn_benchmarks.pyin the default environment.
- Activates the optimized virtualenv and runs
tf_cnn_benchmarks.pyin the Intel Optimized environment.
- Prints the throughput or latency values (for training or inference, respectively) with relative speedups realized with Intel Optimized TensorFlow.
Running the entire benchmarking suite may take some time depending on the hardware you're running on, so please be patient while it is executing.