-
Notifications
You must be signed in to change notification settings - Fork 648
Same training script and model that works in deepmd-kit 3.1.2 fails for 3.1.3 #5350
Hi DP team,
I was trying to use deepmd-kit 3.1.3 on paratera(并行科技)'s HPC service. My task was to finetune the DPA-3.1-3M.pt model (downloaded from ais-square) with the attached input.json script. The same setup worked fine for deepmd-kit 3.1.2 on our group's local HPC, but failed with deepmd-kit 3.1.3 on paratera, with the following error:
Traceback (most recent call last):
File "/data/home/scxj663/run/deepmd-kit/lib/python3.12/configparser.py", line 824, in items
d.update(self._sections[section])
~~~~~~~~~~~~~~^^^^^^^^^
KeyError: 'CONFIG'
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "/data/home/scxj663/run/deepmd-kit/bin/dp", line 8, in
sys.exit(main())
^^^^^^
File "/data/home/scxj663/run/deepmd-kit/lib/python3.12/site-packages/deepmd/main.py", line 1045, in main
deepmd_main = BACKENDSargs.backend.entry_point_hook
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/data/home/scxj663/run/deepmd-kit/lib/python3.12/site-packages/deepmd/backend/pytorch.py", line 66, in entry_point_hook
from deepmd.pt.entrypoints.main import main as deepmd_main
File "/data/home/scxj663/run/deepmd-kit/lib/python3.12/site-packages/deepmd/pt/init.py", line 4, in
from deepmd.pt.cxx_op import (
File "/data/home/scxj663/run/deepmd-kit/lib/python3.12/site-packages/deepmd/pt/cxx_op.py", line 16, in
from deepmd.env import (
File "/data/home/scxj663/run/deepmd-kit/lib/python3.12/site-packages/deepmd/env.py", line 171, in
GLOBAL_CONFIG = _get_package_constants()
^^^^^^^^^^^^^^^^^^^^^^^^
File "/data/home/scxj663/run/deepmd-kit/lib/python3.12/site-packages/deepmd/env.py", line 168, in _get_package_constants
return dict(config.items("CONFIG"))
^^^^^^^^^^^^^^^^^^^^^^
File "/data/home/scxj663/run/deepmd-kit/lib/python3.12/configparser.py", line 827, in items
raise NoSectionError(section)
configparser.NoSectionError: No section: 'CONFIG'
Do you have any idea what's going on? Thank you!
All reactions
Replies: 2 comments 2 replies
BTW I'm currently trying deepmd-3.1.2 on paratera. I'll give feedback later.
All reactions
BTW I'm currently trying deepmd-3.1.2 on paratera. I'll give feedback later.
3.1.2 seemd to work on paratera
All reactions
@dosu Do you have any idea on this issue?
All reactions
The "No section: 'CONFIG'" error indicates that the run_config.ini file is missing or corrupted in your deepmd-kit 3.1.3 installation. This file is generated during the CMake build process and should be located at deepmd/lib/run_config.ini.
There were no functional code changes between 3.1.2 and 3.1.3 that would cause this—PR #4420 only reordered exports alphabetically. So this is likely an installation issue specific to how 3.1.3 was installed on Paratera.
To diagnose, check if the file exists:
python -c "import deepmd; import os; print(os.path.join(deepmd.__path__[0], 'lib/run_config.ini')); print(os.path.exists(os.path.join(deepmd.__path__[0], 'lib/run_config.ini')))"Recommended fixes:
-
Use conda instead of pip (more reliable for HPC with CUDA):
conda create -p /path/to/env python=3.10 deepmd-kit=3.1.3 -c conda-forge
-
Or use the official shell installer from the releases page for your CUDA version
-
Or stick with 3.1.2 if it works for your use case
How did you install deepmd-kit 3.1.3 on Paratera (pip, conda, shell installer)?
To reply, just mention @dosu.
Share context across your team and agents. Try Dosu.
Leave Feedback Learn about deepmd-kit with Dosu Add Dosu to your team