fairchem>=2.0

Fairchem 2 is the current platform for pretrained UMA models, multi-task training, and scalable atomistic inference. It replaces the version 1 trainer, calculator, and model interfaces with a Hydra-configured system built around a shared backbone and task-aware prediction heads. New projects should use fairchem 2 and the current FAIRChemCalculator interface.

We plan to bring back the following models compatible with Fairchem V2 soon:

• Gemnet-OC

• EquiformerV2

• eSEN

Start with the installation guide and Hello World, then use the common workflows for training, evaluation, and scaled inference.

Using Fairchem V1¶

If you need to use models from fairchem version 1, you can still do so by installing version 1:

pip install fairchem-core==1.10

And using the OCPCalculator:

from fairchem.core import OCPCalculator

calc = OCPCalculator(
    model_name="EquiformerV2-31M-S2EF-OC20-All+MD",
    local_cache="pretrained_models",
    cpu=False,
)

Projects and models built on fairchem version v2¶

Projects and models built on fairchem version v1¶