You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Sorry, I seem to have missed this issue as it was posted during the Christmas holidays. The THOR instances should definitely be on the GPU. Is it possible that you haven't started an x-server on your GPU? Doing so is required to be able to run AI2-THOR on the GPU. Assuming you're using a recent version of ai2thor and you have X11 installed (sudo apt install xserver-xorg-core), you can do so by running:
sudo ai2thor-xorg start
This might complain that it can't find the ai2thor-xorg command in which case you can run
which ai2thor-xorg
which should print something like /your/path/to/ai2thor-xorg, you should then copy this path and run
sudo /your/path/to/ai2thor-xorg start
Hello, thank you for your reply. After opening X server,all Thor instances are running on GPU.
Recently, when I was doing one-phase training experiment, I still encountered unitycrashexception, I only started two training processes on one GPU. I found that UnityCrashException errors always occur in the Reset action
The following figure is my training log, with only two vectorsampledtasks.
At present, only Thor instances are running on the GPU, but it always ends after tens of thousands of steps. Can I put the vectorsampledtask: 0, vectorsampledtask: 1 sampling process on the GPU?
The following figure shows the machine parameters in the configuration file
Hello, thank you for your reply. After opening X server,all Thor instances are running on GPU.
Recently, when I was doing one-phase training experiment, I still encountered unitycrashexception, I only started two training processes on one GPU. I found that UnityCrashException errors always occur in the Reset action
The following figure is my training log, with only two vectorsampledtasks.
At present, only Thor instances are running on the GPU, but it always ends after tens of thousands of steps. Can I put the vectorsampledtask: 0, vectorsampledtask: 1 sampling process on the GPU?
The following figure shows the machine parameters in the configuration file
Originally posted by @twb1235 in #29 (comment)
The text was updated successfully, but these errors were encountered: