torchmd-net

torchmd-net copied to clipboard

Cannot create env with mamba

8

I have been trying to use torchmd-net and torchmd-cg. I first tried installing torchmd-net following the instructions given. Tried to create the env with mamba and I encountered the following...

geemi725

Consistent naming between dy and forces

9

This PR renames `y` to `energy` and `dy` to `forces` in places where we mean forces and not gradients. PyTorch-Geometric `Data` objects are now constructed as `Data(energy=, ...)` or `Data(energy=,...

PhilippThoelke

Test loss

2

Which function is supposed to compute the test loss? Can we clean it up? https://github.com/torchmd/torchmd-net/blob/2f6e19091ce161edb86568fe9a2af291a69b38d3/torchmdnet/module.py#L61-L69 Ping: @PhilippThoelke @stefdoerr

raimis

Inconsistency between gradients and forces

9

When training with forces, the model computes the negative gradient of the energy with respect to the positions (a.k.a. forces): https://github.com/torchmd/torchmd-net/blob/db72e12461b1bce9bb3ddebd093fa11a803d37ca/torchmdnet/models/model.py#L196-L207 Some the loaders load forces: https://github.com/torchmd/torchmd-net/blob/db72e12461b1bce9bb3ddebd093fa11a803d37ca/torchmdnet/datasets/ani.py#L279-L281 https://github.com/torchmd/torchmd-net/blob/db72e12461b1bce9bb3ddebd093fa11a803d37ca/torchmdnet/datasets/comp6.py#L100-L102 While the...

raimis

Speed-up neighbors calculation

6

See #61

claudi

Ways to reduce memory use

15

I'm trying to train equivariant transformer models on a GPU with 12 GB of memory. I can train small to medium sized models, but if I make it too large...

peastman

POC: neighbor search

2

This is a proof-of-concept. DO NOT MERGER! - [x] Python wrapper - [x] CPU - [x] Forward pass - [x] CUDA - [x] Forward pass - [x] Backward pass -...

raimis

Periodic boundary conditions

7

I want to try applying a model to a larger system that uses periodic boundary conditions. Currently that isn't supported. It looks to me like the only class that would...

peastman

Optimization of the graph network

14

Optimization of the graph network (`TorchMD_GN`) with NNPOps (https://github.com/openmm/NNPOps). In a special case, `TorchMD_GN` is equivalent to SchNet (https://github.com/compsciencelab/torchmd-net/issues/45#issuecomment-968962455), which is already supported by NNPOps: ```python TorchMD_GN(rbf_type="gauss", trainable_rbf=False, activation="ssp", neighbor_embedding=False)...

raimis

- [x] Integrate the neighbor kernel - [x] Integrate the message kernel - [x] Benchmark Depend on #61 and #69

raimis