[HANDS-ON BUG] mlagents-learn in unit 5 not working

[benbekir]

Describe the bug

The command

!mlagents-learn ./config/ppo/SnowballTarget.yaml --env=./training-envs-executables/linux/SnowballTarget/SnowballTarget --run-id="SnowballTarget1" --no-graphics

doesnt work.
It always results in the following error:

RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu! (when checking argument for argument mat1 in method wrapper_CUDA_addmm).

I have tried specifying the device with the --torch-device argument but that didnt help either.
Maybe this has something to do with the fact that dev versions are used for the ml-agents and ml-agents-envs packages?

  ml-agents: 1.2.0.dev0,
  ml-agents-envs: 1.2.0.dev0,
  Communicator API: 1.5.0,
  PyTorch: 2.5.0+cu121

Material

• Did you use Google Colab?
  Yes

[TPK-MAKG]

Experiencing the same issue. Additionally, the runtime requires a restart after the installation of numpy packages, otherwise it can't find the hyperparameter file in /config/ppo/SnowballTarget.yaml. Maybe it's a package version conflict?

[staffanrolfsson]

I get the same errors.
The first problem after restart (for numpy), is that you will lack the change of working directory "% cd ml-agents", so the following cells will not work as intended. Easy fixed by just adding a cell to do this change and installations following will be correct. But main issue is still there when you try to start the learning:
"RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu! (when checking argument for argument mat1 in method wrapper_CUDA_addmm)"
The Pyramids learning further down in the same Notebook works fine...

[staffanrolfsson]

Tested to change the config file for SnowballTarget, setting threaded: false made the learning possible to start...

[brumocas]

Tested to change the config file for SnowballTarget, setting threaded: false made the learning possible to start...

This worked for me :)