Question Same agent with no changes went from running at 100ish fps to 1-2 fps, any idea why?

I have been tinkering with mlagents for the last week or two. I was able to get a few toy projects working and then it started running incredibly slow. I thought maybe my new agent was messing it up so I went to an old scene: the project built in this video

This one used to, and should, train incredibly fast, but it is a snail now. This leads me to believe it isn't my implementation, but I don't know what it could be. Is there anything that would cause it to run on the CPU instead of the GPU?

A note, it runs at incredible speeds when running off of a brain, it only drops to a snail pace when training so I am guessing it is something with python.

Thank you for any help you can give!

 

You could try to set the torch-device manually. Add --torch-device=cuda to the learn command.