fairchem-core-1.4.0

What’s Changed

New Features / Enhancements

• Add DCP support to main (#938) @rayg1234

Bug Fixes

• OCPCalculator output_only option (#922) @misko
• set tensor dtypes and use torch.tensor to initialize tensors in a2g (#935) @lbluque
• Add cuda set_device for local run (#931) @rayg1234

Dependencies

• pin pytorch at 2_4_0 (#928) @misko

fairchem-core-1.3.0

What’s Changed

New Features / Enhancements

• EquiformerV2 + DeNS model and trainer (#880) @kyonofx
• Add wandb logger init to hydra runners (#894) @rayg1234
• replication use .item(); fix hydra_cli (#920) @kyonofx
• Add utils for logging weight tables and tensor stats (#907) @rayg1234
• Rename hydra heads (#903) @lbluque
• Updated loss and eval metrics (#896) @wood-b
• add hydra freeze backbone option (#898) @misko
• Make eqv2 old to hydra converter work with old old configs :'( (#856) @misko
• Read the logger logging level from environment variable (#874) @misko
• Add script to make release from fine tuned hydra (#875) @misko
• Add hydra entrypoint (#867) @rayg1234
• Add singleton logger (#873) @rayg1234
• add act checkpointing to escn (#852) @misko
• throw error when metadata is non integer (#865) @misko
• change logic to catch missing outputs (#859) @wood-b
• ESCN export II (#848) @rayg1234
• enable ruff formatter (#853) @misko

Bug Fixes

• add ema to BaseTrainer init (#916) @IliasChair
• Sort paths in datasets (#904) @lbluque
• remove destroy_process_group() from finally wrapper as it can hang (#884) @misko
• Add tests for hydra gemnet OC scaling factor generation and loading; Raise error on fail to load scaling factors (#831) @misko
• Match escn_exportable with escn main (#866) @misko
• fix bug where trainer state not being loaded (#863) @rayg1234

Dependencies

• update CI tests to py 3.12 (#858) @lbluque
• bump torch in pyproject.toml (#857) @lbluque

fairchem_data_oc-0.2.0

What’s Changed

Dependencies

• update CI tests to py 3.12 (#858) @lbluque

fairchem_applications_cattsunami-0.2.0

What’s Changed

Documentation

• adding tutorial (#882) @brookwander
• adding link to full dataset (#869) @brookwander

Dependencies

• bump torch in pyproject.toml (#857) @lbluque

fairchem_demo_ocpapi-0.2.0

What’s Changed

Dependencies

• ocpapi dependencies / requests -> 2.32.3 (#862) @kjmichel

fairchem_data_oc-0.1.0

What’s Changed

• Add solvent interface placement code (#765) @mshuaibii
• [BE] Remove large files from fairchem and add references to new location as needed (#761) @levineds
• [BE] Single ruff config (#751) @lbluque
• calc compatability with new configs (#724) @mshuaibii
• Package cleanup (#698) @lbluque
• PyPi CI workflows (#690) @lbluque
• cattsunami packaging (#678) @lbluque
• Fixes and updates to the documentation and website (#676) @zulissimeta
• Consolidate tests into a single folder (#682) @misko
• GA workflows update (#675) @lbluque
• Monorepo (#670) @misko

New Features / Enhancements

• add docstring. Add support for different pseudopotential path (#832) @mshuaibii

Bug Fixes

• explicitly define ispin for oc20 bulks (#840) @mshuaibii
• add tags, fix min_ab slab (#829) @mshuaibii

Deprecations

• Remove large file from repo prior to git history purge (#844) @levineds

fairchem-core-1.2.0

What’s Changed

New Features / Enhancements

• Do not shadow built in property (#841) @lbluque
• add script to port old equiv2 checkpoint+yaml to hydra version (#846) @misko
• Delete distributed option - Always use DDP (#833) @rayg1234
• Torch compile + export escn (#826) @rayg1234
• Add mean e to escn energy head (#828) @rayg1234
• add run group (#827) @rayg1234
• add option to apply mean to energy (#818) @rayg1234
• fix gemnet scaling factors fit.py and add a test (#819) @misko
• Add check to max num atoms (#817) @misko
• Use abs run_dir paths by default (#820) @rayg1234
• FM-v4 branch into main (#752) @misko
• refactor and deprecate old equiformerv2 (#812) @misko
• Fuse all hydras (#814) @rayg1234
• Rank2 tensor head (#792) @lbluque
• Explicitly initialize weights even if initialization method is "uniform" (#809) @misko
• Activation checkpoint equiformersv2 (#811) @rayg1234
• add resolution flag to escn (#804) @misko

Bug Fixes

• Add inference_only option to trainers (#850) @lbluque
• Sort dbs AseDBDataset (#837) @lbluque
• update to gemnet-oc hydra force head to work with amp (#825) @wood-b
• Load normalizers patch (#824) @lbluque
• load linref and normalizer on cuda device when resuming from checkpoint (#813) @misko

fairchem_applications_cattsunami-0.1.0

What’s Changed

• Fix NEB doc notebooks (#773) @brookwander
• [BE] Single ruff config (#751) @lbluque
• calc compatability with new configs (#724) @mshuaibii
• fixing issue with one reaction (#712) @brookwander
• adding seed to md tutorial and ipynb (#710) @brookwander
• Package cleanup (#698) @lbluque
• fixing summary fig and dissociation scheme fig in the gitbook (#700) @brookwander
• PyPi CI workflows (#690) @lbluque
• cattsunami packaging (#678) @lbluque

fairchem_demo_ocpapi-0.1.0

What’s Changed

• [BE] Single ruff config (#751) @lbluque
• no ASE version pin (#739) @lbluque
• move ocpapi to integration tests (#736) @rayg1234
• Package cleanup (#698) @lbluque
• PyPi CI workflows (#690) @lbluque
• Consolidate tests into a single folder (#682) @misko
• GA workflows update (#675) @lbluque
• fix ocpapi test (#681) @misko
• Monorepo (#670) @misko

Dependencies

• Bump requests from 2.31.0 to 2.32.0 in /packages/fairchem-demo-ocpapi (#707) @dependabot

fairchem_core-1.1.0

What’s Changed

• Allow symlink'd experimental folder by using absolute and not resolved paths (#798) @zulissimeta
• fix issues with ddp/hydra and add tests (#796) @misko
• Wrap atom coordinates in ase.Atoms preprocessing (#783) @kyonofx
• Add an option to run PBC in single system mode (#795) @misko
• Move select models to backbone + heads format and add support for hydra (#782) @misko
• [BE] Remove large files from fairchem and add references to new location as needed (#761) @levineds
• clone so3 embedding object (#781) @rayg1234
• add expandable segments var (#775) @rayg1234
• Fix dataset logic (#771) @lbluque
• Relax config (#758) @mshuaibii
• [BE] Single ruff config (#751) @lbluque
• Make relaxation data more general (#714) @anuroopsriram
• Add utils to help run torch profiling (#754) @rayg1234
• Updated ODAC checkpoints & configs (#755) @anuroopsriram
• Hide wandb.watch behind flag (#747) @rayg1234
• Try removing importlib - breaks multiprocess pickling (#746) @rayg1234
• No longer need to specify train dataset loader when predicting (#731) @curtischong
• Log num params to W&B (#741) @rayg1234
• Config fix (#737) @lbluque
• calc compatability with new configs (#724) @mshuaibii
• update preprocess script (#726) @mshuaibii
• support logger entity (#717) @mshuaibii
• pass seed to sampler (#697) @misko

New Features / Enhancements

• Finetune Hydra (#797) @rayg1234
• (OTF) Normalization and element references (#715) @lbluque
• Balanced batch sampler+base dataset (#753) @misko
• add proxy and ssl support. (#778) @haruyuki-oda
• Add slurm qos setting (#757) @rayg1234
• Equiformers2 Graph Parallel (#740) @rayg1234
• Add option to launch distributed runs locally with >1 GPU (#733) @rayg1234

Bug Fixes

• Fix job resume on non-preemption type failures (#803) @rayg1234
• #806: Move data.db to s3 and download via download_large_files (#807) @levineds

Documentation

• #806: Move data.db to s3 and download via download_large_files (#807) @levineds
• [BE] lint/test/docs on PR and push to main (#742) @lbluque

Dependencies

• pin ruff to 0.4.10 (#745) @rayg1234