2.4.1

Hotfix to remove runtime dependency on ase.

Full Changelog: 2.4.0...2.4.1

2.4.0

What's Changed

• MDCATH by @AntonioMirarchi in #337
• Add GLU, Swish, Mish and SwiGLU by @RaulPPelaez in #334
• Add Huber loss, allow choosing training loss function from the yaml by @RaulPPelaez in #335

Full Changelog: 2.3.0...2.4.0

2.3.0

What's Changed

• Disable persistent_workers with num_workers=0 and disable pin_memory in DataModule by @sef43 in #322
• Disable persistent_workers completely, enable pin_memory in DataModule by @RaulPPelaez in #325
• Disable TF32 usage by default by @RaulPPelaez in #327
• Update environment file by @RaulPPelaez in #329
• Disable EMA if the user inputs 0 for alpha by @RaulPPelaez in #330
• Add SPICE 2.0.1, Fix bug in MemmappedDataset by @RaulPPelaez in #333
• MACE-OFF dataset by @RaulPPelaez in #332

New Contributors

• @sef43 made their first contribution in #322

Full Changelog: 2.2.0...2.3.0

2.2.0

What's Changed

• Fix HDF5 not understanding some files by @RaulPPelaez in #313
• Allow to configure the depth of the MLP in output modules by @RaulPPelaez in #314
• torch.compile neighbors without graph breaks by @RaulPPelaez in #305
• Filter non used sample values in QM9 by @RaulPPelaez in #316
• Fix pre-MLP model loading by @RaulPPelaez in #318
• Enable allow_pickle to be able to specify splits. by @brian8128 in #319

New Contributors

• @brian8128 made their first contribution in #319

Full Changelog: 2.1.0...2.2.0

2.1.0

What's Changed

• add arg to MemmappedDataset by @AntonioMirarchi in #295
• Add custom training loop to the docs by @RaulPPelaez in #298
• Add SPICE 1.1.4, remove leftover remove_ref_energy in some datasets by @RaulPPelaez in #303
• Added support for ensemble models by @stefdoerr in #308
• Support for ensemble model zip files by @stefdoerr in #311

Full Changelog: 2.0.0...2.1.0

2.0.0

This release marks the first stable release of TorchMD-Net.

Version 2.0 accompanies the publication https://arxiv.org/abs/2402.17660 .

What's Changed

• Use switching function for Coulomb prior by @peastman in #287
• Update README.md by @guillemsimeon in #294

Full Changelog: 0.15.2...2.0.0

0.15.2

Another minor hotfix for CI tests.
Full Changelog: 0.15.1...0.15.2

0.15.1

Minor fix for test 1bbb03a .

Full Changelog: 0.15.0...0.15.1

0.15.0

What's Changed

• Add OpenMM integration example by @RaulPPelaez in #272
• Float precision of energies by @stefdoerr in #275
• Better handling of prior configuration by @RaulPPelaez in #280
• Clean up some datasets by @RaulPPelaez in #281
• Training with relative energies, infering with total energy. by @RaulPPelaez in #279
• Mark standardize as deprecated by @RaulPPelaez in #278
• Send atomref to correct device during training by @RaulPPelaez in #284
• Make TorchMD_Net always return two tensors by @RaulPPelaez in #283
• Fix EnergyRefRemover by @stefdoerr in #285
• Learnable Atomref by @RaulPPelaez in #282
• update ANI-2x training configuration by @stefdoerr in #286
• Add dtype parameter to External by @RaulPPelaez in #274

This release breaks compatibility with checkpoints trained with versions previous to 0.2.5. If you really need to load such a checkpoint with this version onward, you can try to patch this in:
6abe174

Full Changelog: 0.14.2...0.15.0

0.14.2

What's Changed

• Small fix for OSX by @RaulPPelaez in #270

Full Changelog: 0.14.1...0.14.2