0.5.9

• Add frozen-weight finetuning option + wizard
• add shift:auto option for finetuning
• add SMAX models to wizard
• some bugfixes with import

Full Changelog: 0.5.8...0.5.9

0.5.8

• add SMAX models
• add fitting of general tensors (gen_tensor)
• add 2L parallelization export
• update tutorials

What's Changed

• Release/0.5.8 by @yury-lysogorskiy in #27

Full Changelog: 0.5.7.2...0.5.8

0.5.7.2

Upd license to ASL 0.5.7.2

Release 0.5.7

Bug fix for the presets config

Full Changelog: 0.5.6...0.5.7

Release 0.5.6

What's Changed

• update GRACE-1L/2L presets according to latest foundational models (with suffix _latest)

• update input file template with new recommended optimizer settings

• version 0.5.6 by @yury-lysogorskiy in #21

Full Changelog: 0.5.5...0.5.6

0.5.5

• bugfix reduce_elements functionality when elements are not ordered lexicographically
• add _configure_keras_backend to the top-most module

What's Changed

• Release 0.5.5 by @yury-lysogorskiy in #19

Full Changelog: 0.5.4...0.5.5

0.5.4

Windows/MacOS support

What's Changed

• Feature/windows by @yury-lysogorskiy in #16

Full Changelog: 0.5.3...0.5.4

0.5.3

Add support of extxyz. Some bugfixes

What's Changed

• Enable reading extxyz directly in input.yaml by @pmrv in #12
• bugfix: elements reduce in BondSpecificRadialBasisFunction

New Contributors

• @pmrv made their first contribution in #12

Full Changelog: 0.5.2...0.5.3

0.5.2

Update to 0.5.2:

add more foundational UMLIPS (GRACE-1L/2L-OMAT/OAM-medium/large): checkpoints and saved models
add grace_utils::resave_checkpoint
grace_predict: add stress prediction
grace_preprocess: add EQUI_NEIGH_STRATEGY, add FULL_ELEMENTS preset, add --remove_stage1 arg, add support of cutoff dict
upd documentation
asecalculator.py::TPCalculator: change defaults for padding pad_neighbors_fraction: 0.25 -> 0.05, pad_atoms_number=10->1
gracemaker.py:
-- add dataset plotting
-- add more learning rate schedulers, store epoch into checkpoint
instructions: add MLPRadialFunction_v2, InvariantLayerRMSNorm and TrainableShiftTarget

What's Changed

• Release/0.5.2 by @yury-lysogorskiy in #11

Full Changelog: 0.5.1...0.5.2

0.5.1

Update to 0.5.1: Fine-tuning of foundation models

• gracemaker: add functionality for finetuning of foundation models
• add GRACE-*-OMAT foundation models and checkpoints for finetuning
• add grace_utils for model conversion, exporting and summary
• README.md: add important notes for model conversion
• implementation details: using dict-of-instructions instead of list-of-instructions