Lack of GPU support on the platform

NeurIPS 2019 : MineRL Competition
#1

Hello, I think your server does not support a GPU, even using "gpu": true in aicrowd.json. I’m using pytorch==1.2 I checked two scenarios:

  • create the model, and call model.cuda() - the evaluation stops after the first iteration. Create the model without moving it to cuda(using cpu) - it works fine.
  • create the model and load locally pre-trained weights(on the GPU) - the evaluation stops after the first iteration. I think this is also due to the lack of a GPU in your environment. In my local environment with the same docker image everything works fine.
1 Like
#2

Also, about the second scenario. I made another submit (by the way, I lost a lot of my submits to find the reason of stopping the evaluation). Everything works fine using torch.load (weights_path, map_location = 'cpu') - loading pre-trained weights directly into the cpu.

1 Like