Core bus dumped error when train wav2vec2 model

Hello, @patrickvonplaten @valhalla
I’m fine-tuning wav2vec model with Fine-Tune XLSR-Wav2Vec2 for low-resource ASR with :hugs: Transformers by using Chinese common voice dataset at local machine with 4XA40 Nvidia. The memory is 46068MiB.
I have some problems with training.
I used the tutorial to train on Colab with another customer dataset is fine but not the Chinese common Voice. So I used it to train on a remote GPU (4 by A40) with docker setting.
Every time when it reached to train the model. It always crashed by core bus dumped error. All of my GPU does not look like running at all. I don’t know why Colab can run but the local can’t run and I didn’t change any code. I tried my code on Juputer notebook and terminal. Both of them has core dumped error.

Do I need add any code to let it access gpu or can the model run on multiple gpu?