Open maniast opened 2 years ago
I may have encountered a similar issue. Have you found any way to solve it? Which task are you training?
having the same issue. All AMP samples crashing on start
some common errors
ed: ../../../source/plugins/carb/gym/impl/Gym/GymPhysX.cpp: 5993 [Error] [carb.gym.plugin] Gym cuda error: an illegal memory access was encountered: ../../../source/plugins/carb/gym/impl/Gym/GymPhysXCuda.cu: 937 [Error] [carb.gym.plugin] Failed to fill root state tenso
Solution is to decrease batch size , contact nums, env_nums. Since these vatiables' memory allocation consuming required pytorch's memory allocation
I find the problem in a while circle.
So I stopped the circle after first implement,then I got another previous error:
This is caused by implementgym.simulate(sim)
before gym.prepare_sim(sim)
,I added it before the while circle containing
gym.simulate(sim)
,the problem was resolved
I keep getting this at random time periods during my training. The specific error:
/buildAgent/work/45f70df4210b2e3e/source/gpunarrowphase/src/PxgNarrowphaseCore.cpp (9908) : internal error : GPU compressContactStage1 fail to launch kernel stage 1!!
/buildAgent/work/45f70df4210b2e3e/source/gpunarrowphase/src/PxgNarrowphaseCore.cpp (9945) : internal error : GPU compressContactStage2 fail to launch kernel stage 1!!