Closed newplay closed 1 year ago
The input file looks fine. From the error message, it seems that the error has something to do with pytorch. There is even no other message in standard output before the error, so probably the error occurs when trying to import pytorch at the very beginning of the training process. If this is the case, you might want to double check your system environment.
In addition, we have never tested DeepH-E3 with pytorch 2.0, so maybe you can try using lower versions of pytorch (for example, 1.9.0). If this does not work, you might also try replacing pytorch geometric with version 1.7.2 and e3nn with version 0.3.5.
Update: it is reported that DeepH-E3 works fine with pytorch 2.0, e3nn 0.4.4 and pytorch-geometric 2.2.
Thank you for your response. After adjusting the environment settings as you suggested, the issue has been resolved. I am extremely grateful! However, after testing, using pytorch=2.1.0+cu121, e3nn=5.1.0, and torch_geometric=2.3.1 has shown better efficiency, reducing the time per epoch from 500s to 360s.
Good to know that your problem is solved! It is also good news that upgrading the environment will make the training a lot faster.
In training processed , I got the error of
Segmentation fault (core dumped)
, there is mytrain.ini
below: #################################################################################################################################################################### and the version : #################################################################################################
the device of my computer is: ram 64G GPU RTX3070 CPU amd-R5-5600X I have no idea what's happening. So I run the gdb and get this information: