| source ~/.bashrc |
| conda activate mace-scf |
|
|
| mpirun -np 16 python /home/jhm/software/mace-tools/scripts/will_train.py \ |
| --name="mace_pol_spice_v1" \ |
| --train_file="../SPICE_2_dipeptides_all_configs_train.xyz" \ |
| --test_file "../SPICE_2_dipeptides_all_configs_test.xyz" \ |
| --E0s="{35: -70045.28385080204, 6: -1030.5671648271828, 17: -12522.649269035726, 9: -2715.318528602957, 1: -13.571964772646918, 53: -8102.524593409054, 7: -1486.3750255780376, 8: -2043.933693071156, 15: -9287.407133426237, 16: -10834.4844708122}"\ |
| --config_type_weights='{"Default":1.0}' \ |
| --model="Polarizable" \ |
| --hidden_irreps='64x0e + 64x1o' \ |
| --loss="energy_forces_atomic_multipoles" \ |
| --distributed \ |
| --num_workers=4\ |
| --num_interactions=2 \ |
| --r_max=5.0 \ |
| --energy_weight=40\ |
| --eval_interval=2 \ |
| --batch_size=32 \ |
| --valid_batch_size=32 \ |
| --max_num_epochs=800\ |
| --error_table="DensityEnergyRMSE" \ |
| --swa\ |
| --start_swa=600\ |
| --device="cuda"\ |
| --kspace_cutoff_factor=1.0 \ |
| --field_dependence_type="biased_local_linear" \ |
| --energy_key="energy" \ |
| --forces_key="forces" \ |
| --atomic_multipoles_weight=10 \ |
| --swa_atomic_multipoles_weight=1000 \ |
| --atomic_multipoles_max_l=0 \ |
| --atomic_multipoles_smearing_width=1.5 \ |
| --keep_checkpoints \ |
| --atomic_multipoles_key="mbis_multipoles" \ |
| --default_dtype="float64" \ |
| --restart_latest \ |
| --scf_training_options='{"num_scf_steps":10, "constant_charge":True, "mixing_parameter": 1.0, "use_autograd_forces": True}' |
|
|
