Hi, Is it possible to check whether rllib using tensorflow gpu or not? I try to run ray tune with PPO Algorithm and num_gpus = 1, but It only using 1GB /16GB of my GPU. Is this because the rllib using tensorflow CPU instead of GPU?
cc @mfitton
RLlib will raise an error if GPU support isn't working actually. The reason it's only using 1GB is since your model/batch size is small. For RL, it's very uncommon to use lots of GPU memory since most models are quite small.
Most helpful comment
RLlib will raise an error if GPU support isn't working actually. The reason it's only using 1GB is since your model/batch size is small. For RL, it's very uncommon to use lots of GPU memory since most models are quite small.