“died with <Signals.SIGSEGV: 11>.” when evaluate

[ppbangKGT]

Hi, thanks for your nice work! When I learned your code and evaluate the validation dataset of openlane, there was an error:
subprocess.CalledProcessError: Command '['/home/pbw/anaconda3/envs/lanemd_torch18/bin/python3', '-u', 'main_persformer.py', '--local_rank=1', '--mod=PersFormer', '--batch_size=8']' died with <Signals.SIGSEGV: 11>.
Here is the screenshot:

[ilnehc]

@ppbangKGT Maybe you could check https://github.com/OpenPerceptionX/PersFormer_3DLane/issues/16 if it is the same problem concerning of CUDA environment.

[ppbangKGT]

@ppbangKGT Maybe you could check #16 if it is the same problem concerning of CUDA environment.

Actually, I don't meet the CUDA_HOME error. And if i run the code like #8 python main_persformer.py --mod=PersFormer --batch_size=1, there will be an error "Segmentation fault".

[ilnehc]

Did you try to use 1 GPU (launch with --nproc_per_node 1)?

In similar issues, others said "Signals.SIGSEGV: 11 seems to be a common bug when you are using different libraries/environments for compiling and running the code."

Could you post your environment information and driver version? A helpful script could be collect_env.

[ppbangKGT]

Did you try to use 1 GPU (launch with --nproc_per_node 1)?

In similar issues, others said "Signals.SIGSEGV: 11 seems to be a common bug when you are using different libraries/environments for compiling and running the code."

Could you post your environment information and driver version? A helpful script could be collect_env.

I think this error may be caused by the version of cuda. When I use another server whose CUDA version is 11.3. The problem is solved. Thanks for your time.