Dipole fitting on multi-GPU

[VondrakMar]

I have tried to train a model with --model="AtomicDipolesMACE", and --loss="dipole" with multiple GPUs. This gave me error below. However, training normal MACE on energy and forces worked normally.

Error I got:

RuntimeError: Expected to have finished reduction in the prior iteration before starting a new one. This error indicates that your module has parameters that were not used in producing loss. You can enable unused parameter detection by passing the keyword argument `find_unused_parameters=True` to `torch.nn.parallel.DistributedDataParallel`, and by 
making sure all `forward` function outputs participate in calculating loss. 
If you already have done the above, then the distributed data parallel module wasn't able to locate the output tensors in the return value of your module's `forward` function. Please include the loss function and the structure of the return value of `forward` of your module when reporting this issue (e.g. list, dict, iterable).
Parameter indices which did not receive grad for rank 0: 8 9 10 11 12 13 14 22 23 24 25
 In addition, you can set the environment variable TORCH_DISTRIBUTED_DEBUG to either INFO or DETAIL to print out information about which particular parameters did not receive gradient on this rank as part of this error 
    if torch.is_grad_enabled() and self.reducer._rebuild_buckets():
RuntimeError: Expected to have finished reduction in the prior iteration before starting a new one. This error indicates that your module has parameters that were not used in producing loss. You can enable unused parameter detection by passing the keyword argument `find_unused_parameters=True` to `torch.nn.parallel.DistributedDataParallel`, and by 
making sure all `forward` function outputs participate in calculating loss. 
If you already have done the above, then the distributed data parallel module wasn't able to locate the output tensors in the return value of your module's `forward` function. Please include the loss function and the structure of the return value of `forward` of your module when reporting this issue (e.g. list, dict, iterable).
Parameter indices which did not receive grad for rank 1: 8 9 10 11 12 13 14 22 23 24 25
 In addition, you can set the environment variable TORCH_DISTRIBUTED_DEBUG to either INFO or DETAIL to print out information about which particular parameters did not receive gradient on this rank as part of this error 
srun: error: ravg1193: task 0: Exited with exit code 1 

My dipole training input (this did not work on multi-gpu, but is normally running with 1 GPU):

srun mace_run_train \
    --name="MACE_model" \
    --train_file="pubchem.xyz" \
    --atomic_numbers="[1,2, 5,6, 7, 8, 9,14, 15, 16,17, 18, 35, 53]" \
    --config_type_weights='{"Default":1.0}' \
    --model="AtomicDipolesMACE" \
    --distributed \
    --E0s="average" \
    --hidden_irreps='128x0e + 128x1o' \
    --r_max=5.0 \
    --batch_size=10 \
    --valid_batch_size=5\
    --max_num_epochs=100 \
    --restart_latest \
    --device=cuda \
    --loss="dipole"\
    --dipole_key="scf_dipole"\
    --error_table="DipoleRMSE"\
    --forces_key="dft_forces"\
    --energy_key="dft_energy"

E+F training with same dataset (works)

srun mace_run_train \
    --name="MACE_model" \
    --train_file="pubchem_smoll.xyz" \
    --atomic_numbers="[1,2, 5,6, 7, 8, 9,14, 15, 16,17, 18, 35, 53]" \
    --config_type_weights='{"Default":1.0}' \
    --distributed \
    --E0s="average" \
    --hidden_irreps='128x0e + 128x1o' \
    --r_max=5.0 \
    --batch_size=10 \
    --valid_batch_size=5\
    --max_num_epochs=100 \
    --restart_latest \
    --device=cuda \
    --dipole_key="scf_dipole"\
    --forces_key="dft_forces"\
    --energy_key="dft_energy"