fairchem_data_omol-0.1.1

What’s Changed

New Features / Enhancements

• omol lbfgs evals (#1488) @mshuaibii

fairchem-core-2.5.0

What’s Changed

New Features / Enhancements

• Lammps Integration v0 (#1454) @rayg1234
• add oc25 (#1490) @misko
• omol lbfgs evals (#1488) @mshuaibii
• Add slurm_additional_parameters to cli submitit submission (#1419) @zulissimeta

Bug Fixes

• Fix attribute device and dtype for conversion to ASE Atoms (#1340) @recisic

Deprecations

• Deprecate tensorboard logger (#1482) @lbluque

Dependencies

• revert clusterscope (#1469) @misko
• Update lmdb version to 1.7.3 (#1418) @zulissimeta

fairchem-core-2.4.0

What’s Changed

New Features / Enhancements

• Add OMol evaluations (#1412) @mshuaibii
• Inference server + client for running local MP graph parallel (#1415) @rayg1234
• [ BE ] refactored init edge rot mat (#1368) @misko
• [ BE ] add user accessible embeddings (#1421) @misko
• diatomic tests (#1429) @wood-b
• add stress migration option (#1241) @misko
• enable updating the edges of an AtomicData batch (#1397) @arturtoshev
• Add inference-only tasks to allow predicting on tasks not trained on (#1411) @rayg1234
• support eval restarts (#1367) @mshuaibii
• Load tasks from checkpoint when running benchmarks (#1312) @lbluque
• bump clusterscope (#1376) @misko
• add fp64 support (#1337) @misko

Bug Fixes

• StrEnum backwards compatibility (#1428) @lbluque
• Modify clamp in radius_graph_pbc (#1303) @ericyuan00000
• Filter annoying warning (#1413) @rayg1234
• Fix fine-tuning configs (#1406) @jagritisahoo
• fix: correct expected value in external_graph_gen error message (True… (#1347) @zulissimeta
• Fix a bug in AtomicData.to_ase_single (#1310) (#1346) @zulissimeta
• revert lmdb version (#1351) @zulissimeta
• upgrading clusterscope (#1345) @luccabb
• if-condition requires pbc is None, but below we index pbc (#1309) (#1314) @arturtoshev

Documentation

• fix docs, reset calc (#1447) @mshuaibii

Tests

• added simple so2 tests, impossible vectors test for escn_md, and typehints (#1414) @bkmi

Dependencies

• Make pymatgen an optional dependency (#1445) @lbluque
• fix ase version; latest ase is breaking quacc/book build (#1426) @misko
• revert lmdb version (#1351) @zulissimeta
• upgrading clusterscope (#1345) @luccabb

fairchem_data_omol-0.1.0

What’s Changed

New Features / Enhancements

• Add release workflows for fairchem-data-omol (#1443) @lbluque
• omdata cleanup (#1437) @mshuaibii

fairchem_data_oc-1.0.2

What’s Changed

New Features / Enhancements

• ncore=4 (#1381) @mshuaibii

Documentation

• more style docs tweaks & typos (#1238) @zulissimeta

fairchem_applications_cattsunami-1.1.1

What’s Changed

Documentation

• more style docs tweaks & typos (#1238) @zulissimeta

fairchem-core-2.3.0

What’s Changed

Release of UMA-s 1.1 (uma-s-1p1) and UMA-m 1.1 (uma-m-1p1) checkpoints as well as the arxiv paper (https://arxiv.org/abs/2506.23971).

Major changes

• UMA-m is the best in class model for all tasks across the board.
• Both Version 1.1 checkpoints fixes a composition bug which made relaxations and simulations on molecular crystals unstable in some cases and violated extensivity property for periodic systems. This fix also improved the accuracy of omol task over v1.0.
• We also turned off TF32 by default (users can still manually turned it back on for speed) as we were observing it reduced the accuracy of a few property predictions that were highly sensitive to numerical precision (for phonon calculations for example). This will change your results slightly if you upgrade to 2.3.0 and make inference slower, you can switch it back manually by setting tf32=True in the inference_settings.
• A new simple finetuning procedure has been added.

New Features / Enhancements

• UMA 1.1 checkpoint release (#1321) @mshuaibii
• UMA Finetune scripts and docs (#1248) @mshuaibii
• Dont use tf32 on by default (#1283) @rayg1234
• Add option to choose cache directory where model is saved (#1264) @suttergustavo
• adding clusterscope (#1265) @luccabb
• add option to return checkpoint along with model (#1243) @lbluque
• assert task datasets is subset of backbone datasets (#1263) @lbluque
• release isolated atom refs (#1185) @mshuaibii
• Remove mlip_unit specific logic from train_runner, add Checkpointable protocol (#1233) @rayg1234

Bug Fixes

• Add osc -> omc migration for non ema checkpoint (#1286) @rayg1234
• Fix Preemption Checkpointing Mechamism (#1242) @ryanliu30
• add in model_version to mole and fix composition embebdding (#1258) @misko
• fix setting implemented properties (#1239) @lbluque

Documentation

• update FAIRChemCalculator initialization in benchmark configs (#1254) @lbluque
• more style docs tweaks & typos (#1238) @zulissimeta

fairchem_data_oc-1.0.1

What’s Changed

New Features / Enhancements

• tile solvents (#1106) @mshuaibii

fairchem-core-2.2.0

What’s Changed

This version enables batch inference and a variety of bug fixes for the calculator/predictor inference, as well as new documentation

New Features / Enhancements

• Batch inference cleanup (#1181) @lbluque
• FAIRChem calculator from_model_checkpoint classmethod (#1203) @lbluque
• allow automatic device setting when loading mlip predict units (#1202) @lbluque
• Enable local runs (#1172) @rayg1234

Bug Fixes

• add pbc v2 grid resolution bounds (#1204) @misko
• Fix bug when taskname is enum instead of str (#1223) @rayg1234
• Fix ac mole bug (#1205) @rayg1234

Documentation

• Docs cleanup & molecular info add (#1206) @zulissimeta
• version uma (#1199) @mshuaibii

fairchem_applications_cattsunami-1.1.0

What’s Changed

Removed OCPNEB to enable fairchemv2 support

Documentation

• Docs cleanup & molecular info add (#1206) @zulissimeta
• Fairchemv2 docs update (#1160) @rayg1234