首先感谢您非常好的工作。
我复现的过程中,按照Train TPVFormer for lidar segmentation task on 3090 with 24G GPU memory.的步骤,四卡GPU训练时可以正常训练,但是使用八卡3090时会出现“_pickle.UnpicklingError: pickle data was truncated”的问题,请问是哪里出了问题呢
We run into the same error when we set num_workers for dataloader to a value larger than one. And when we used 0 or 1 for num_workers, this problem went away.
首先感谢您非常好的工作。 我复现的过程中,按照Train TPVFormer for lidar segmentation task on 3090 with 24G GPU memory.的步骤,四卡GPU训练时可以正常训练,但是使用八卡3090时会出现“_pickle.UnpicklingError: pickle data was truncated”的问题,请问是哪里出了问题呢