Add FlashMD #21
Add FlashMD #21
Conversation
There was a problem hiding this comment.
Can you remove directory from git?
|
|
||
| kokkosable = 1; | ||
| execution_space = ExecutionSpaceFromDevice<DeviceType>::space; | ||
| atomKK = (AtomKokkos *) atom; |
|
|
||
| #ifdef FIX_CLASS | ||
| // clang-format off | ||
| FixStyle(flashmd,FixFlashMD); |
There was a problem hiding this comment.
So as mentioned already I don't think we should call this fix flashmd, unless we want to use FlashMD as the name for the generic method of generating positions and not the specific model. I also feel like this should contain metatomic in the name, like the pair style.
Couple of ideas: fix metatomic_pm, fix metatomic_pos_vel, fix metatomic_ml_dyn. I don't really love any of these though, so we should keep thinking about it!
|
|
||
| namespace LAMMPS_NS { | ||
| class MetatomicSystemAdaptor; | ||
| class PairMetatomicData; |
There was a problem hiding this comment.
This should be a separate FixFlashMDData/Fix<WhateverNameWeEndUpWith>Data class, so we can evolve the two parts separately.
| std::vector<torch::DeviceType> available_devices(); | ||
| void pick_device(torch::Device* device, const char* requested); | ||
| void initial_integrate(int) override; | ||
| void init_list(int id, NeighList *ptr) override; | ||
|
|
||
| protected: |
There was a problem hiding this comment.
| std::vector<torch::DeviceType> available_devices(); | |
| void pick_device(torch::Device* device, const char* requested); | |
| void initial_integrate(int) override; | |
| void init_list(int id, NeighList *ptr) override; | |
| protected: | |
| void initial_integrate(int) override; | |
| void init_list(int id, NeighList *ptr) override; | |
| protected: | |
| std::vector<torch::DeviceType> available_devices(); | |
| void pick_device(torch::Device* device, const char* requested); |
just to make it clear which functions are part of the fix API or not
| double *mass = atom->mass; | ||
| int *type = atom->type; | ||
| int *mask = atom->mask; | ||
| if (igroup == atom->firstgroup) nlocal = atom->nfirst; |
There was a problem hiding this comment.
| if (igroup == atom->firstgroup) nlocal = atom->nfirst; | |
| if (igroup == atom->firstgroup) { | |
| nlocal = atom->nfirst; | |
| } |
| int nghost = atom->nghost; | ||
| int nall = nlocal + nghost; | ||
|
|
||
| double *mass = atom->mass; |
There was a problem hiding this comment.
what's the difference between rmass and mass?
|
|
||
| auto properties = metatensor_torch::LabelsHolder::single()->to(mta_data->device); | ||
| auto block = torch::make_intrusive<metatensor_torch::TensorBlockHolder>( | ||
| // TODO: is there a way to check what dtype the model expects for input data? |
There was a problem hiding this comment.
yes, model->capabilities().dtype
| ); | ||
| mta_data->evaluation_options->set_selected_atoms(selected_atoms); | ||
|
|
||
| // call the model to get delta-positions and updated momenta |
There was a problem hiding this comment.
| // call the model to get delta-positions and updated momenta | |
| // call the model to get new positions and momenta |
| v[i][0] = momenta[i][0].item<double>(); | ||
| v[i][1] = momenta[i][1].item<double>(); | ||
| v[i][2] = momenta[i][2].item<double>(); |
There was a problem hiding this comment.
shouldn't these be divided by the masses?
Add a new fix that uses a model to predict future positions and momenta.