Change Log

2.1.3

  • Fixed an incorrect message-passing convention in the PyG and DGL TensorNet interaction and embedding blocks. Edge messages are now aggregated onto the source (center) node so that each atom correctly collects information from its neighbors. Pre-trained PyG TensorNet weights have been re-released to match the corrected convention. (#758, @kenko911)
  • Refreshed the PyG TensorNet README, added a missing TrajectoryObserver, improved PESCalculator stress-unit handling and logging, and tightened the related unit tests. (#758, @kenko911)
  • Reverted the experimental PyG ports of QET, QETWarp, and the PyG electrostatics modules pending further validation; the DGL QET model remains the supported implementation.

2.1.2

  • Added Hugging Face Hub support for loading pre-trained models, with automatic fallback checking and respect for the MATGL_CACHE environment variable.
  • Removed the deprecated hubconf.py (superseded by Hugging Face support).
  • Added TensorNetWrapper integrating the NVIDIA nvalchemi-toolkit for fully GPU-resident MD/Relax workflows, including an example script for NVT MD. (#754)
  • Added an example notebook for training a QET potential with PyTorch Lightning. (#742)
  • Added the QET-MatQ-PES FP dataset with accompanying README. (#743)
  • Added documentation on typical MatGL installation time for the DGL backend. (#741)
  • Bumped supported torch upper bound to <=2.11.0. (#746)
  • Upgraded pymatgen-core and refreshed uv.lock.
  • Test suite cleanup and improved coverage for IO utilities.

2.1.1

  • Merged TensorNet (PyG) and TensorNetWarp into a single TensorNet class with optional warp acceleration (use_warp parameter; auto-detected when nvalchemi-toolkit-ops is installed).
  • Moved warp-accelerated TensorEmbedding and TensorNetInteraction layers to matgl.layers._embedding_warp and matgl.layers._graph_convolution_warp.
  • Made nvalchemiops an optional dependency throughout: _pymatgen_pyg, _ase_pyg, and warp layer imports all fall back gracefully to pymatgen-based neighbor list construction when the package is absent.

2.1.0

  • Bug fix for accidental change of default backend.
  • Training module updated for QET support.

2.0.9

  • Bug fix for missing Atoms2Graph export.

2.0.7

  • Refactored PyG TensorNet embedding and interaction blocks to pure PyTorch for improved compatibility. (@kenko911)
  • Improved handling of stress units in PESCalculator. (@kenko911)
  • Enabled returning intermediate crystal features from CHGNet and TensorNet models. (@bowen-bd)
  • Added GPU-accelerated neighbor list construction and improved CUDA neighbor list performance and retry logic. (@zubatyuk)
  • Integrated NVIDIA TensorNet Warp CUDA kernels into the main branch. (@atulcthakur, @zubatyuk)
  • Improved QET training support via updates to Atoms2Graph, collate_fn_pes, and MGLDataset (including include_ref_charge). (@kenko911)
  • Documentation updates for QET, including references and DOI links. (@kenko911)

2.0.6

  • Bug fix for CHGnet loading.

2.0.5

  • Improved error messages for backend/model mismatch. Try to transparently handle simple situations.

2.0.4

  • Bug fix for matgl.graph.data and matgl.graph.converter imports for different backends.

2.0.3

  • Bug fix for matgl.ext.pymatgen import for different backends.

2.0.2

  • QET (Charge-Equilibrated TensorNet) architecture and pre-trained weights are added!
  • Begun a migration to Pytorch-Geometric over the now-deprecated DGL. So far, only vanilla TensorNet has been implemented in PYG). DGL models still work but require a manual setup (change of backend and installation of DGL).

1.2.7

  • Use original custom RemoteFile rather than fsspec, which is very finicky with SSL connections.
  • _create_directed_line_graph error handling (@bowen-bd)
  • Update Import Alias for lightning (@jcwang587)
  • Add nvt_nose_hoover to MD ensemble (@bowen-bd)
  • Allow training of magmom when no line graph presents (@bowen-bd)
  • Allow disable BondGraph in CHGNet (@bowen-bd)

1.2.6

  • Fix missing torchdata dependency for Linux.

1.2.5

  • Dependency pinning now is platform specific. Linux based systems can now work with latest DGL and torch.

1.2.3

  • Fix dependency issues with DGL. Pinning to DGL<=2.1.0 for now, which have versions for all OSes.

1.2.1

  • Bug fix for pbc dtype on Windows systems.

1.2.0

  • Release of MatPES-based models.
  • Pin DGL and PyTorch dependencies to 2.2.0 to ensure compatibility with Mac.

1.1.3

  • Improve the memory efficiency and speed of three-body interactions. (@kenko911)
  • FrechetCellFilter is added for variable cell relaxation in Relaxer class. (@kenko911)
  • Smooth l1 loss function is added for training. (@kenko911)

1.1.2

  • Move AtomRef Fitting to numpy to avoid bug (@BowenD-UCB)
  • NVE ensemble added (@kenko911)
  • Migrate from pytorch_lightning to lightning.

1.1.1

  • Pin dependencies to support latest DGL 2.x. @kenko911

1.1.0

  • Implementation of CHGnet + pre-trained models. (@BowenD-UCB)

1.0.0

  • First 1.0.0 release to reflect the maturity of the matgl code! All changes below are the efforts of @kenko911.
  • Equivariant TensorNet and SO3Net are now implemented in MatGL.
  • Refactoring of M3GNetCalculator and M3GNetDataset into generic PESCalculator and MGLDataset for use with all models instead of just M3GNet.
  • Training framework has been unified for all models.
  • ZBL repulsive potentials has been implemented.

0.9.2

  • Added Tensor Placement Calls For Ease of Training with PyTorch Lightning (@melo-gonzo).
  • Allow extraction of intermediate outputs in “embedding”, “gc_1”, “gc_2”, “gc_3”, and “readout” layers for use as atom, bond, and structure features. (@JiQi535)

0.9.1

  • Update Potential version numbers.

0.9.0

  • set pbc_offsift and pos as float64 by @lbluque in https://github.com/materialsvirtuallab/matgl/pull/153
  • Bump pytorch-lightning from 2.0.7 to 2.0.8 by @dependabot in https://github.com/materialsvirtuallab/matgl/pull/155
  • add cpu() to avoid crash when using ase with GPU by @kenko911 in https://github.com/materialsvirtuallab/matgl/pull/156
  • Added the united test for hessian in test_ase.py to improve the coverage score by @kenko911 in https://github.com/materialsvirtuallab/matgl/pull/157
  • AtomRef Updates by @lbluque in https://github.com/materialsvirtuallab/matgl/pull/158
  • Bump pymatgen from 2023.8.10 to 2023.9.2 by @dependabot in https://github.com/materialsvirtuallab/matgl/pull/160
  • Remove torch.unique for finding the maximum three body index and little cleanup in united tests by @kenko911 in https://github.com/materialsvirtuallab/matgl/pull/161
  • Bump pymatgen from 2023.9.2 to 2023.9.10 by @dependabot in https://github.com/materialsvirtuallab/matgl/pull/162
  • Add united test for trainer.test and description in the example by @kenko911 in https://github.com/materialsvirtuallab/matgl/pull/165
  • Bump pytorch-lightning from 2.0.8 to 2.0.9 by @dependabot in https://github.com/materialsvirtuallab/matgl/pull/167
  • Sequence instead of list for inputs by @lbluque in https://github.com/materialsvirtuallab/matgl/pull/169
  • Avoiding crashes for PES training without stresses and update pretrained models by @kenko911 in https://github.com/materialsvirtuallab/matgl/pull/168
  • Bump pymatgen from 2023.9.10 to 2023.9.25 by @dependabot in https://github.com/materialsvirtuallab/matgl/pull/173
  • Allow to choose distribution in xavier_init by @lbluque in https://github.com/materialsvirtuallab/matgl/pull/174
  • An example for the simple training of M3GNet formation energy model is added by @kenko911 in https://github.com/materialsvirtuallab/matgl/pull/176
  • Directed line graph by @lbluque in https://github.com/materialsvirtuallab/matgl/pull/178
  • Bump pymatgen from 2023.9.25 to 2023.10.4 by @dependabot in https://github.com/materialsvirtuallab/matgl/pull/180
  • Bump torch from 2.0.1 to 2.1.0 by @dependabot in https://github.com/materialsvirtuallab/matgl/pull/181
  • Bump pymatgen from 2023.10.4 to 2023.10.11 by @dependabot in https://github.com/materialsvirtuallab/matgl/pull/183
  • add testing to m3gnet potential training example by @lbluque in https://github.com/materialsvirtuallab/matgl/pull/179
  • Update Training a MEGNet Formation Energy Model with PyTorch Lightning by @1152041831 in https://github.com/materialsvirtuallab/matgl/pull/185
  • Bump pymatgen from 2023.10.11 to 2023.11.12 by @dependabot in https://github.com/materialsvirtuallab/matgl/pull/187
  • dEdLat contribution for stress calculations is added and Universal Potentials are updated by @kenko911 in https://github.com/materialsvirtuallab/matgl/pull/189
  • Bump torch from 2.1.0 to 2.1.1 by @dependabot in https://github.com/materialsvirtuallab/matgl/pull/190

New Contributors

  • @1152041831 made their first contribution in https://github.com/materialsvirtuallab/matgl/pull/185

Full Changelog: https://github.com/materialsvirtuallab/matgl/compare/v0.8.5…v0.8.6

0.8.3

  • Extend the functionality of ASE-interface for molecular systems and include more different ensembles. (@kenko911)
  • Improve the dgl graph construction and fix the if statements for stress and atomwise training. (@kenko911)
  • Refactored MEGNetDataset and M3GNetDataset classes with optimizations.

0.8.5

  • Bug fix for np.meshgrid. (@kenko911)

0.8.2

  • Add site-wise predictions for Potential. (@lbluque)
  • Enable CLI tool to be used for multi-fidelity models. (@kenko911)
  • Minor fix for model version for DIRECT model.

0.8.1

  • Fixed bug with loading of models trained with GPUs.
  • Updated default model for relaxations to be the M3GNet-MP-2021.2.8-DIRECT-PES model.

0.8.0

  • Fix a bug with use of set2set in M3Gnet implementation that affected intensive models such as the formation energy model. M3GNet model version is updated to 2 to invalidate previous models. Note that PES models are unaffected. (@kenko911)

0.7.1

  • Minor optimizations for memory and isolated atom training (@kenko911)

0.7.0

  • MatGL now supports structures with isolated atoms. (@JiQi535)
  • Fourier expansion layer and generalize cutoff polynomial. (@lbluque)
  • Radial bessel (zeroth order bessel). (@lbluque)

0.6.2

  • Simple CLI tool mgl added.

0.6.1

  • Bug fix for training loss_fn.

0.6.0

  • Refactoring of training utilities. Added example for training an M3GNet potential.

0.5.6

  • Minor internal refactoring of basis expansions into _basis.py. (@lbluque)

0.5.5

  • Critical bug fix for code regression affecting pre-loaded models.

0.5.4

  • M3GNet Formation energy model added, with example notebook.
  • M3GNet.predict_structure method added.
  • Massively improved documentation at http://matgl.ai.

0.5.3

  • Minor doc and code usability improvements.

0.5.2

  • Minor improvements to model versioning scheme.
  • Added matgl.get_available_pretrained_models() to help with model discovery.
  • Misc doc and error message improvements.

0.5.1

  • Model versioning scheme implemented.
  • Added convenience method to clear cache.

0.5.0

  • Model serialization has been completely rewritten to make it easier to use models out of the box.
  • Convenience method matgl.load_model is now the default way to load models.
  • Added a TransformedTargetModel.
  • Enable serialization of Potential.
  • IMPORTANT: Pre-trained models have been reserialized. These models can only be used with v0.5.0+!

0.4.0

  • Pre-trained M3GNet universal potential
  • Pytorch lightning training utility.

v0.3.0

  • Major refactoring of MEGNet and M3GNet models and organization of internal implementations. Only key API are exposed via matgl.models or matgl.layers to hide internal implementations (which may change).
  • Pre-trained models ported over to new implementation.
  • Model download now implemented.

v0.2.1

  • Fixes for pre-trained model download.
  • Speed up M3GNet 3-body computations.

v0.2.0

  • Pre-trained MEGNet models for formation energies and band gaps are now available.
  • MEGNet model implemented with predict_structure convenience method.
  • Example notebook demonstrating pre-trained model usage is available.

v0.1.0

  • Initial working version with m3gnet and megnet.
© Copyright 2022, Materials Virtual Lab
© Copyright 2022, Materials Virtual Lab