DriveML Huawei Autonomous Vehicles Challenge Forum

Go back to competition Back to thread list Post in this thread

> Could we use GPU to accelerate the training?How to use GPU in ray?

As the title

Posted by: Team85 @ Nov. 8, 2019, 4:21 a.m.

Take a look at the Ray and RLlib documentation:
https://ray.readthedocs.io/en/latest/rllib-training.html
https://ray.readthedocs.io/en/latest/using-ray-with-gpus.html

You can of course use GPUs, however, bear in mind that the benefits of doing so depend on the number of parameters of the neural network you use for training, which will in turn depend on the kinds of input data your model receives. CPU training often has an advantage when network size is small, for example. Hybrid approaches are also viable, see e.g.: https://arxiv.org/abs/1611.06256

Posted by: HuaweiUK @ Nov. 8, 2019, 11:17 a.m.
Post in this thread