Tensormismatch restore_experiment

[PurerStress]

Hello,

I have a problem regarding restoring my experiment.
When I do experiment_config.restore_file in my python file, I get the error:

RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu!

I previously trained my model on the GPU and sampled and buffered on the CPU. When I reload the model I want to keep it this way but then I get this error. If I reload my experiment an train it on the CPU it works fine.
Is there any way to solve this problem?

[matteobettini]

Hey, have you tried playing with this?

BenchMARL/benchmarl/conf/experiment/base_experiment.yaml

Line 122 in 567acd9

restore_map_location: null

[PurerStress]

Thank you for your response.
I did try it according to the documentation, but I was unable to map it right.
I tried it with:
experiment_config.restore_map_location = {"cuda:0": "cpu"} which again gave me the error and if I map it like
experiment_config.restore_map_location = {"cpu": "cuda:0"} it makes the collection time much slower.
Do I miss something?

[Xmaster6y]

restore_map_location="cpu" works for me