OTF SGP models with force_only=True, not working with OTF.from_checkpint()

[usaleem97]

Hi FLARE team,

I am training a SGP potential with force_only tag set to True. If I try resuming training by using the from_checkpoint() method, it spits out the following errors:

Traceback (most recent call last):
File "restart.py", line 30, in
test_otf= OTF.from_checkpoint('otf_checkpt.json')
File "/project/ssharada_53/usaleem/flare/flare/learners/otf.py", line 968, in from_checkpoint
otf_model = cls.from_dict(json.loads(f.readline()))
File "/project/ssharada_53/usaleem/flare/flare/learners/otf.py", line 920, in from_dict
flare_calc, _kernels = SGP_Calculator.from_file(dct["flare_calc"])
File "/project/ssharada_53/usaleem/flare/flare/bffs/sgp/calculator.py", line 164, in from_file
sgp, kernels = SGP_Wrapper.from_dict(gp_dict["gp_model"])
File "/project/ssharada_53/usaleem/flare/flare/bffs/sgp/sparse_gp.py", line 295, in from_dict
struc_stress = train_struc.stress
File "/project/ssharada_53/usaleem/flare/flare/atoms.py", line 102, in stress
return self.get_stress()
File "/home1/usaleem/.conda/envs/flare/lib/python3.8/site-packages/ase/atoms.py", line 829, in get_stress
assert shape == (6,)
AssertionError

I did a short OTF SGP trajectory on the same system using force_only=False, and it seemed to work without a problem.

I have attached my run and restart script
script.docx
script.docx

Thank you in advance!
Usama
~

[anjohan]

Hi Usama,

Thank you for your interest in FLARE!

Which version of FLARE are you using?

I don't see the struc_stress = train_struc.stress line anymore, and we fixed some issues related to stress training in #343.

[usamaS97]

Hi Anders,

Thanks for your response. I am on the 1.4.1 version. And the issue is with the version I am using again.