Train ERROR: single GPU is OK,but multi-GPU is wrong

594422814 / UDT_pytorch

MIT License

191 stars 23 forks source link

Train ERROR: single GPU is OK,but multi-GPU is wrong #3

Open Hevans123 opened 5 years ago

Hevans123 commented 5 years ago

Hello, If I use one GPU,I can run train_UDT.py successfully, but when using multi-GPUs, it went wrong. I use pytorch=1.0.0 The information is like this RuntimeError: CUDA error: an illegal memory access was encountered (copy_device_to_device at /pytorch/aten/src/ATen/native/cuda/Copy.cu:98)

zyc08 commented 4 years ago

Hello,have you solved this question?I meet the same error.

EliSchwartz commented 4 years ago

same here