Tensorflow comes with example code to retrain Inception's final layer for new image categories. The code is found here [url removed, login to view] and the description here [url removed, login to view]
The code in [url removed, login to view] is only able to use one GPU even on a machine with several GPUs. I need it to be modified to be able to use all the GPUs on a machine in order to accelerate the training. The modified script must also accept a flag to have it use a specific number of GPUs.