Question [Solve]RuntimeError: The size of tensor a (#) must match the size of tensor b (#) at non-singleton..

RuntimeError: The size of tensor a (3) must match the size of tensor b (15) at non-singleton dimension 1
^ in cmd promt not editor, as the very last message as it abruptly stops

This error started occuring after I set observation stacks to 5 for the camera sensor of my agent.
The training is fine for aproximately 10k steps before this error occurs and it exports the results and shuts down.
Does anyone know what's going on here?

 

same problem

 

[Solved] I initialized from a new run-id, and although I'm unsure if this helped too, but I made "observable attribute handling" in behavior parameters to examine all.

 

are you taking already trained model and building more trainning over that? like --iniitalize-from=""
If this is the case then this occurs due to incompatibility with the previous trained model. Like if you have changed vector observation size or ray perception count etc.
So if you make it same with the previous model then error will remove