Stress is not fitted/printed out

[park2971]

I am a beginner user of the mace software. I have an issue about fitting stress data.

I am trying a fine-tuning with:
mace_run_train
--name="test_SrSnO3"
--energy_key="energy"
--forces_key="forces"
--stress_key="stress"
--E0s='{38:-0.03202952, 50:-0.07540552, 8:-0.06285466}'
--foundation_model="small"
--multiheads_finetuning=False
--train_file="train.xyz"
--valid_fraction=0.05
--energy_weight=1.0
--forces_weight=10
--compute_stress=True
--stress_weight=1.0
--lr=0.01
--scaling="rms_forces_scaling"
--batch_size=2
--max_num_epochs=6
--swa
--swa_energy_weight=100
--swa_forces_weight=10
--swa_stress_weight=1.0
--start_swa=4
--ema
--ema_decay=0.99
--amsgrad
--default_dtype="float64"
--device=cuda
--wandb
--wandb_project="$WANDB_PROJECT"
--wandb_name="$WANDB_NAME"
--seed=3 > log_mace_train

I believe my tarin.xyz file has a correct form and it looks like:
20
Lattice="8.126034319 0.0 0.0 0.0 8.126034319 0.0 0.0 0.0 4.063017159" Properties=species:S:1:pos:R:3:forces:R:3 stress="2.0671878224925348e-05 -0.0 0.0 -0.0 2.0671878224925348e-05 -0.0 0.0 -0.0 2.08778480260795e-05" free_energy=-132.04269078 energy=-132.04269078 pbc="T T T"
Sr 0.00000000 0.00000000 0.00000000 -0.00000000 0.00000000 0.00000000
Sr 0.00000000 4.06302000 0.00000000 -0.00000000 0.00000000 0.00000000
Sr 4.06302000 0.00000000 0.00000000 -0.00000000 0.00000000 0.00000000
Sr 4.06302000 4.06302000 0.00000000 -0.00000000 0.00000000 0.00000000
Sn 2.03151000 2.03151000 2.03151000 0.00000000 0.00000000 -0.00000000
Sn 2.03151000 6.09453000 2.03151000 0.00000000 0.00000000 0.00000000
Sn 6.09453000 2.03151000 2.03151000 -0.00000000 -0.00000000 0.00000000
Sn 6.09453000 6.09453000 2.03151000 -0.00000000 0.00000000 -0.00000000
O 2.03151000 2.03151000 0.00000000 0.00000000 -0.00000000 0.00000000
O 2.03151000 6.09453000 0.00000000 -0.00000000 -0.00000000 0.00000000
O 6.09453000 2.03151000 0.00000000 -0.00000000 -0.00000000 0.00000000
O 6.09453000 6.09453000 0.00000000 0.00000000 -0.00000000 0.00000000
O 2.03151000 0.00000000 2.03151000 0.00000000 -0.00000000 -0.00000000
O 2.03151000 4.06302000 2.03151000 -0.00000000 -0.00000000 0.00000000
O 6.09453000 0.00000000 2.03151000 -0.00000000 0.00000000 0.00000000
O 6.09453000 4.06302000 2.03151000 0.00000000 -0.00000000 0.00000000
O 0.00000000 2.03151000 2.03151000 0.00000000 0.00000000 -0.00000000
O 0.00000000 6.09453000 2.03151000 0.00000000 0.00000000 -0.00000000
O 4.06302000 2.03151000 2.03151000 0.00000000 -0.00000000 0.00000000
O 4.06302000 6.09453000 2.03151000 0.00000000 0.00000000 -0.00000000
...

Even though I set compute_stress and stress_weight above, the log and results do not show any information about stress.

The below is log file:

cuequivariance or cuequivariance_torch is not available. Cuequivariance acceleration will be disabled.
2025-11-16 23:57:39.084 INFO: ===========VERIFYING SETTINGS===========
2025-11-16 23:57:39.084 INFO: Stage Two is activated as start_stage_two was defined
2025-11-16 23:57:39.084 INFO: MACE version: 0.3.15
2025-11-16 23:57:39.283 INFO: CUDA version: 12.6, CUDA device: 0
2025-11-16 23:57:39.827 INFO: Using foundation model mace-mp-0 small as initial checkpoint.
Using Materials Project MACE for MACECalculator with /users/7/park2971/.cache/mace/20231210mace128L0_energy_epoch249model
Using float64 for MACECalculator, which is slower but more accurate. Recommended for geometry optimization.
2025-11-16 23:57:40.430 INFO: CUDA version: 12.6, CUDA device: 0
2025-11-16 23:57:40.453 INFO: Using head Default out of ['Default']
2025-11-16 23:57:40.453 INFO: ===========LOADING INPUT DATA===========
2025-11-16 23:57:40.453 INFO: Using heads: ['Default']
2025-11-16 23:57:40.453 INFO: Using the key specifications to parse data:
2025-11-16 23:57:40.453 INFO: Default: KeySpecification(info_keys={'energy': 'energy', 'stress': 'stress', 'virials': 'REF_virials', 'dipole': 'dipole', 'head': 'head', 'elec_temp': 'elec_temp', 'total_charge': 'total_charge', 'polarizability': 'polarizability', 'total_spin': 'total_spin'}, arrays_keys={'forces': 'forces', 'charges': 'REF_charges'})
2025-11-16 23:57:40.453 INFO: ============= Processing head Default ===========
2025-11-16 23:57:40.811 WARNING: Since ASE version 3.23.0b1, using energy_key 'energy' is no longer safe when communicating between MACE and ASE. We recommend using a different key, rewriting 'energy' to 'REF_energy'. You need to use --energy_key='REF_energy' to specify the chosen key name.
2025-11-16 23:57:40.884 WARNING: Since ASE version 3.23.0b1, using forces_key 'forces' is no longer safe when communicating between MACE and ASE. We recommend using a different key, rewriting 'forces' to 'REF_forces'. You need to use --forces_key='REF_forces' to specify the chosen key name.
2025-11-16 23:57:40.960 WARNING: Since ASE version 3.23.0b1, using stress_key 'stress' is no longer safe when communicating between MACE and ASE. We recommend using a different key, rewriting 'stress' to 'REF_stress'. You need to use --stress_key='REF_stress' to specify the chosen key name.
2025-11-16 23:57:41.055 INFO: Training set 1/1 [energy: 575, stress: 575, virials: 0, dipole components: 0, head: 575, elec_temp: 0, total_charge: 0, polarizability: 0, total_spin: 0, forces: 575, charges: 0]
2025-11-16 23:57:41.058 INFO: Total Training set [energy: 575, stress: 575, virials: 0, dipole components: 0, head: 575, elec_temp: 0, total_charge: 0, polarizability: 0, total_spin: 0, forces: 575, charges: 0]
2025-11-16 23:57:41.058 INFO: No validation set provided, splitting training data instead.
2025-11-16 23:57:41.063 INFO: Using random 5% of training set for validation with indices saved in: ./test_SrSnO3_valid_indices_3.txt
2025-11-16 23:57:41.066 INFO: Random Split Training set [energy: 547, stress: 547, virials: 0, dipole components: 0, head: 547, elec_temp: 0, total_charge: 0, polarizability: 0, total_spin: 0, forces: 547, charges: 0]
2025-11-16 23:57:41.066 INFO: Random Split Validation set [energy: 28, stress: 28, virials: 0, dipole components: 0, head: 28, elec_temp: 0, total_charge: 0, polarizability: 0, total_spin: 0, forces: 28, charges: 0]
2025-11-16 23:57:41.066 INFO: Total number of configurations: train=547, valid=28, tests=[],
2025-11-16 23:57:41.072 INFO: Atomic Numbers used: [8, 38, 50]
2025-11-16 23:57:41.072 INFO: Isolated Atomic Energies (E0s) not in training file, using command line argument
2025-11-16 23:57:41.072 INFO: Atomic Energies used (z: eV) for head Default: {8: -0.06285466, 38: -0.03202952, 50: -0.07540552}
2025-11-16 23:57:41.072 INFO: Processing datasets for head 'Default'
2025-11-16 23:57:43.209 INFO: Combining 1 list datasets for head 'Default'
2025-11-16 23:57:43.282 INFO: Head 'Default' training dataset size: 547
2025-11-16 23:57:43.283 INFO: Computing average number of neighbors
2025-11-16 23:57:43.950 INFO: Average number of neighbors: 66.68699090549624
2025-11-16 23:57:43.950 INFO: During training the following quantities will be reported: energy, forces
2025-11-16 23:57:43.950 INFO: ===========MODEL DETAILS===========
2025-11-16 23:57:44.465 INFO: Loading FOUNDATION model
2025-11-16 23:57:44.466 INFO: Using filtered elements: [8, 38, 50]
2025-11-16 23:57:44.467 INFO: Model configuration extracted from foundation model
2025-11-16 23:57:44.467 INFO: Using weighted loss function for fine-tuning
2025-11-16 23:57:44.467 INFO: Message passing with hidden irreps 128x0e)
2025-11-16 23:57:44.467 INFO: 2 layers, each with correlation order: 3 (body order: 4) and spherical harmonics up to: l=3
2025-11-16 23:57:44.467 INFO: Radial cutoff: 6.0 A (total receptive field for each atom: 12.0 A)
2025-11-16 23:57:44.467 INFO: Distance transform for radial basis functions: None
2025-11-16 23:57:45.706 INFO: Total number of parameters: 402202
2025-11-16 23:57:45.706 INFO:
2025-11-16 23:57:45.706 INFO: ===========OPTIMIZER INFORMATION===========
2025-11-16 23:57:45.706 INFO: Using ADAM as parameter optimizer
2025-11-16 23:57:45.706 INFO: Batch size: 2
2025-11-16 23:57:45.706 INFO: Using Exponential Moving Average with decay: 0.99
2025-11-16 23:57:45.706 INFO: Number of gradient updates: 1641
2025-11-16 23:57:45.706 INFO: Learning rate: 0.01, weight decay: 5e-07
2025-11-16 23:57:45.706 INFO: WeightedEnergyForcesLoss(energy_weight=1.000, forces_weight=10.000)
2025-11-16 23:57:45.707 INFO: Stage Two (after 4 epochs) with loss function: WeightedEnergyForcesLoss(energy_weight=100.000, forces_weight=10.000), with energy weight : 100.0, forces weight : 10.0 and learning rate : 0.001
2025-11-16 23:57:45.786 INFO: Using Weights and Biases for logging
2025-11-16 23:58:01.514 INFO: Using gradient clipping with tolerance=10.000
2025-11-16 23:58:01.514 INFO:
2025-11-16 23:58:01.514 INFO: ===========TRAINING===========
2025-11-16 23:58:01.514 INFO: Started training, reporting errors on validation set
2025-11-16 23:58:01.514 INFO: Loss metrics on validation set
2025-11-16 23:58:04.865 INFO: Initial: head: Default, loss=4.37995880, RMSE_E_per_atom= 6383.41 meV, RMSE_F= 108.79 meV / A
2025-11-16 23:58:48.500 INFO: Epoch 0: head: Default, loss=0.00336150, RMSE_E_per_atom= 54.11 meV, RMSE_F= 56.47 meV / A
2025-11-16 23:59:28.624 INFO: Epoch 1: head: Default, loss=0.00272838, RMSE_E_per_atom= 42.16 meV, RMSE_F= 51.44 meV / A
2025-11-17 00:00:07.818 INFO: Epoch 2: head: Default, loss=0.00189164, RMSE_E_per_atom= 33.57 meV, RMSE_F= 43.05 meV / A
2025-11-17 00:00:46.746 INFO: Epoch 3: head: Default, loss=0.00135831, RMSE_E_per_atom= 30.63 meV, RMSE_F= 36.15 meV / A
2025-11-17 00:00:46.806 INFO: Changing loss based on Stage Two Weights
2025-11-17 00:01:25.671 INFO: Epoch 4: head: Default, loss=0.00146093, RMSE_E_per_atom= 4.83 meV, RMSE_F= 35.27 meV / A
2025-11-17 00:02:05.291 INFO: Epoch 5: head: Default, loss=0.00110682, RMSE_E_per_atom= 1.43 meV, RMSE_F= 33.39 meV / A
2025-11-17 00:02:05.342 INFO: Training complete
2025-11-17 00:02:05.343 INFO:
2025-11-17 00:02:05.343 INFO: ===========RESULTS===========
2025-11-17 00:02:05.348 INFO: Loading checkpoint: ./checkpoints/test_SrSnO3_run-3_epoch-3.pt
2025-11-17 00:02:05.405 INFO: Loaded Stage one model from epoch 3 for evaluation
2025-11-17 00:02:05.406 INFO: Saving model to checkpoints/test_SrSnO3_run-3.model
2025-11-17 00:02:05.555 INFO: Compiling model, saving metadata to test_SrSnO3_compiled.model
2025-11-17 00:02:06.131 INFO: Computing metrics for training, validation, and test sets
2025-11-17 00:02:06.133 INFO: Skipping evaluation for heads: ['pt_head']
2025-11-17 00:02:06.133 INFO: Evaluating train_Default ...
2025-11-17 00:02:19.899 INFO: Evaluating valid_Default ...
2025-11-17 00:02:20.455 INFO: Error-table on TRAIN and VALID:
+---------------+---------------------+------------------+-------------------+
| config_type | RMSE E / meV / atom | RMSE F / meV / A | relative F RMSE % |
+---------------+---------------------+------------------+-------------------+
| train_Default | 28.9 | 54.4 | 3.06 |
| valid_Default | 30.6 | 36.2 | 3.08 |
+---------------+---------------------+------------------+-------------------+
2025-11-17 00:02:40.190 INFO: Loading checkpoint: ./checkpoints/test_SrSnO3_run-3_epoch-5_swa.pt
2025-11-17 00:02:40.250 INFO: Loaded Stage two model from epoch 5 for evaluation
2025-11-17 00:02:40.250 INFO: Saving model to checkpoints/test_SrSnO3_run-3_stagetwo.model
2025-11-17 00:02:40.423 INFO: Compiling model, saving metadata test_SrSnO3_stagetwo_compiled.model
2025-11-17 00:02:40.878 INFO: Computing metrics for training, validation, and test sets
2025-11-17 00:02:40.879 INFO: Skipping evaluation for heads: ['pt_head']
2025-11-17 00:02:40.880 INFO: Evaluating train_Default ...
2025-11-17 00:02:54.694 INFO: Evaluating valid_Default ...
2025-11-17 00:02:55.210 INFO: Error-table on TRAIN and VALID:
+---------------+---------------------+------------------+-------------------+
| config_type | RMSE E / meV / atom | RMSE F / meV / A | relative F RMSE % |
+---------------+---------------------+------------------+-------------------+
| train_Default | 1.9 | 45.7 | 2.56 |
| valid_Default | 1.4 | 33.4 | 2.85 |
+---------------+---------------------+------------------+-------------------+
2025-11-17 00:03:14.396 INFO: Done

The below is slurm.out file:

/users/7/park2971/lib/python3.12/site-packages/e3nn/o3/_wigner.py:10: UserWarning: Environment variable TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD detected, since theweights_only argument was not explicitly passed to torch.load, forcing weights_only=False.
_Jd, _W3j_flat, _W3j_indices = torch.load(os.path.join(os.path.dirname(file), 'constants.pt'))
/users/7/park2971/lib/python3.12/site-packages/mace/calculators/mace.py:197: UserWarning: Environment variable TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD detected, since theweights_only argument was not explicitly passed to torch.load, forcing weights_only=False.
torch.load(f=model_path, map_location=device)
/users/7/park2971/lib/python3.12/site-packages/torch/jit/_check.py:178: UserWarning: The TorchScript type system doesn't support instance-level annotations on empty non-base types in __init__. Instead, either 1) use a type annotation in the class body, or 2) wrap the type in torch.jit.Attribute.
warnings.warn(
/users/7/park2971/lib/python3.12/site-packages/torch/jit/_check.py:178: UserWarning: The TorchScript type system doesn't support instance-level annotations on empty non-base types in __init__. Instead, either 1) use a type annotation in the class body, or 2) wrap the type in torch.jit.Attribute.
warnings.warn(
/users/7/park2971/lib/python3.12/site-packages/torch/jit/_check.py:178: UserWarning: The TorchScript type system doesn't support instance-level annotations on empty non-base types in __init__. Instead, either 1) use a type annotation in the class body, or 2) wrap the type in torch.jit.Attribute.
warnings.warn(
/users/7/park2971/lib/python3.12/site-packages/torch/jit/_check.py:178: UserWarning: The TorchScript type system doesn't support instance-level annotations on empty non-base types in __init__. Instead, either 1) use a type annotation in the class body, or 2) wrap the type in torch.jit.Attribute.
warnings.warn(
/users/7/park2971/lib/python3.12/site-packages/torch/jit/_check.py:178: UserWarning: The TorchScript type system doesn't support instance-level annotations on empty non-base types in __init__. Instead, either 1) use a type annotation in the class body, or 2) wrap the type in torch.jit.Attribute.
warnings.warn(
/users/7/park2971/lib/python3.12/site-packages/torch/jit/_check.py:178: UserWarning: The TorchScript type system doesn't support instance-level annotations on empty non-base types in __init__. Instead, either 1) use a type annotation in the class body, or 2) wrap the type in torch.jit.Attribute.
warnings.warn(
/users/7/park2971/lib/python3.12/site-packages/torch/jit/_check.py:178: UserWarning: The TorchScript type system doesn't support instance-level annotations on empty non-base types in __init__. Instead, either 1) use a type annotation in the class body, or 2) wrap the type in torch.jit.Attribute.
warnings.warn(
/users/7/park2971/lib/python3.12/site-packages/torch/jit/_check.py:178: UserWarning: The TorchScript type system doesn't support instance-level annotations on empty non-base types in __init__. Instead, either 1) use a type annotation in the class body, or 2) wrap the type in torch.jit.Attribute.
warnings.warn(
/users/7/park2971/lib/python3.12/site-packages/torch/jit/_check.py:178: UserWarning: The TorchScript type system doesn't support instance-level annotations on empty non-base types in __init__. Instead, either 1) use a type annotation in the class body, or 2) wrap the type in torch.jit.Attribute.
warnings.warn(
/users/7/park2971/lib/python3.12/site-packages/torch/jit/_check.py:178: UserWarning: The TorchScript type system doesn't support instance-level annotations on empty non-base types in __init__. Instead, either 1) use a type annotation in the class body, or 2) wrap the type in torch.jit.Attribute.
warnings.warn(
/users/7/park2971/lib/python3.12/site-packages/torch/jit/_check.py:178: UserWarning: The TorchScript type system doesn't support instance-level annotations on empty non-base types in __init__. Instead, either 1) use a type annotation in the class body, or 2) wrap the type in torch.jit.Attribute.
warnings.warn(
/users/7/park2971/lib/python3.12/site-packages/torch/jit/_check.py:178: UserWarning: The TorchScript type system doesn't support instance-level annotations on empty non-base types in __init__. Instead, either 1) use a type annotation in the class body, or 2) wrap the type in torch.jit.Attribute.
warnings.warn(
/users/7/park2971/lib/python3.12/site-packages/torch/jit/_check.py:178: UserWarning: The TorchScript type system doesn't support instance-level annotations on empty non-base types in __init__. Instead, either 1) use a type annotation in the class body, or 2) wrap the type in torch.jit.Attribute.
warnings.warn(
/users/7/park2971/lib/python3.12/site-packages/torch/jit/_check.py:178: UserWarning: The TorchScript type system doesn't support instance-level annotations on empty non-base types in __init__. Instead, either 1) use a type annotation in the class body, or 2) wrap the type in torch.jit.Attribute.
... ( about wandb) ...
/users/7/park2971/lib/python3.12/site-packages/mace/tools/checkpoint.py:187: UserWarning: Environment variable TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD detected, since theweights_only argument was not explicitly passed to torch.load, forcing weights_only=False.
torch.load(f=checkpoint_info.path, map_location=device),
/users/7/park2971/lib/python3.12/site-packages/torch/jit/_check.py:178: UserWarning: The TorchScript type system doesn't support instance-level annotations on empty non-base types in __init__. Instead, either 1) use a type annotation in the class body, or 2) wrap the type in torch.jit.Attribute.
warnings.warn(
/users/7/park2971/lib/python3.12/site-packages/mace/tools/checkpoint.py:187: UserWarning: Environment variable TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD detected, since theweights_only argument was not explicitly passed to torch.load, forcing weights_only=False.
torch.load(f=checkpoint_info.path, map_location=device),
/users/7/park2971/lib/python3.12/site-packages/torch/jit/_check.py:178: UserWarning: The TorchScript type system doesn't support instance-level annotations on empty non-base types in __init__. Instead, either 1) use a type annotation in the class body, or 2) wrap the type in torch.jit.Attribute.
warnings.warn(

Can anyone give me an advice for this?

[park2971]

Stress is fitted now after adding:
--loss "stress"

[ilyes319]

You can use "stress" loss or "Universal" loss.