Models¶
The machine-learned interatomic potentials (MLIPs) known to ML-PEG are listed
below, with the configuration defined in ml_peg/models/models.yml. Models
marked (not currently enabled) are commented out in models.yml and are not
run in the current benchmarks; they are listed here for completeness.
Note
Citations and longer descriptions are still to be added for each model.
MACE¶
mace-mp-0a¶
mace-mp-0a:
module: mace.calculators
class_name: mace_mp
device: "auto"
trained_on_dispersion: false
level_of_theory: PBE
kwargs:
model: "medium"
mace-mp-0b3¶
mace-mp-0b3:
module: mace.calculators
class_name: mace_mp
device: "auto"
trained_on_dispersion: false
level_of_theory: PBE
kwargs:
model: "medium-0b3"
mace-mpa-0¶
mace-mpa-0:
module: mace.calculators
class_name: mace_mp
device: "auto"
trained_on_dispersion: false
level_of_theory: PBE
kwargs:
model: "medium-mpa-0"
mace-omat-0¶
mace-omat-0:
module: mace.calculators
class_name: mace_mp
device: "auto"
trained_on_dispersion: false
level_of_theory: PBE
kwargs:
model: "medium-omat-0"
mace-matpes-r2scan¶
mace-matpes-r2scan:
module: mace.calculators
class_name: mace_mp
device: "auto"
trained_on_dispersion: false
level_of_theory: r2SCAN
kwargs:
model: "mace-matpes-r2scan-0"
head: "matpes_pbe"
dispersion_kwargs:
xc: r2scan
label: D4
mace-mh-1-omat¶
mace-mh-1-omat:
module: mace.calculators
class_name: mace_mp
device: "auto"
trained_on_dispersion: false
level_of_theory: PBE
kwargs:
model: "mh-1"
head: omat_pbe
MACE-OFF23(L)¶
MACE-OFF23(L):
module: mace.calculators
class_name: mace_off
device: "auto"
trained_on_dispersion: true
kwargs:
model: large
MACE-OFF24(M)¶
MACE-OFF24(M):
module: mace.calculators
class_name: mace_off
device: "auto"
trained_on_dispersion: true
kwargs:
model: "https://raw.githubusercontent.com/ACEsuit/mace-off/main/mace_off24/MACE-OFF24_medium.model"
mace-omol¶
mace-omol:
module: mace.calculators
class_name: mace_omol
device: "auto"
trained_on_dispersion: true
level_of_theory: ωB97M-V/def2-TZVPD
kwargs:
model: "extra_large"
mace-mh-1-omol¶
mace-mh-1-omol:
module: mace.calculators
class_name: mace_mp
device: "auto"
trained_on_dispersion: true
level_of_theory: ωB97M-V/def2-TZVPD
kwargs:
model: "mh-1"
head: omol
mace-polar-1-s¶
mace-polar-1-s:
module: mace.calculators
class_name: mace_polar
device: "cpu"
trained_on_dispersion: true
level_of_theory: ωB97M-V
kwargs:
model: "polar-1-s"
mace-polar-1-m¶
mace-polar-1-m:
module: mace.calculators
class_name: mace_polar
device: "cpu"
trained_on_dispersion: true
level_of_theory: ωB97M-V
kwargs:
model: "polar-1-m"
mace-polar-1-l¶
mace-polar-1-l:
module: mace.calculators
class_name: mace_polar
device: "cpu"
trained_on_dispersion: true
level_of_theory: ωB97M-V
kwargs:
model: "polar-1-l"
DPA¶
Install these fixed-float32 DeePMD checkpoints with uv sync --extra dpa.
DPA-3.3-1M uses OpenLAM-v1 with the OMat24 head and the DPA4 variants use OMat24.
DPA-3.3-1M is CC-BY-4.0, while DPA4 is CC-BY-NC-4.0 and restricted to
non-commercial use.
dpa-3p3-1M-omat¶
dpa-3p3-1M-omat:
module: deepmd.calculator
class_name: DP
datasets: [OpenLAM-v1, OMAT, MPtrj, OC20, OC22, ODAC23, SPICE2]
trained_on_dispersion: false
level_of_theory: PBE
kwargs:
model: DPA-3.3-1M
head: Omat24
dpa-4-nano-omat¶
dpa-4-nano-omat:
module: deepmd.calculator
class_name: DP
datasets: [OMAT]
trained_on_dispersion: false
level_of_theory: PBE
kwargs:
model: DPA4-Nano-OMat24-v20260805
dpa-4-neo-omat¶
dpa-4-neo-omat:
module: deepmd.calculator
class_name: DP
datasets: [OMAT]
trained_on_dispersion: false
level_of_theory: PBE
kwargs:
model: DPA4-Neo-OMat24-v20260805
dpa-4-plus-omat¶
dpa-4-plus-omat:
module: deepmd.calculator
class_name: DP
datasets: [OMAT]
trained_on_dispersion: false
level_of_theory: PBE
kwargs:
model: DPA4-Plus-OMat24-v20260805
Orb¶
orb-v3-consv-inf-omat¶
orb-v3-consv-inf-omat:
module: orb_models.inference.calculator
class_name: OrbCalc
device: "cpu"
trained_on_dispersion: false
level_of_theory: PBE
kwargs:
name: "orb_v3_conservative_inf_omat"
orb-v3-direct-inf-omat¶
orb-v3-direct-inf-omat:
module: orb_models.inference.calculator
class_name: OrbCalc
device: "cpu"
trained_on_dispersion: false
level_of_theory: PBE
kwargs:
name: "orb_v3_direct_inf_omat"
orb-v3-consv-omol¶
orb-v3-consv-omol:
module: orb_models.forcefield.inference.calculator
class_name: OrbCalc
device: "cpu"
trained_on_dispersion: true
level_of_theory: PBE
kwargs:
name: "orb_v3_conservative_omol"
PET¶
pet-mad¶
pet-mad:
module: upet.calculator
class_name: UPETCalculator
device: "cpu"
trained_on_dispersion: false
level_of_theory: PBEsol
kwargs:
model: "pet-mad-s"
version: "1.0.2"
dispersion_kwargs:
xc: pbesol
pet-mad-s-1p5¶
pet-mad-s-1p5:
module: upet.calculator
class_name: UPETCalculator
device: "cpu"
trained_on_dispersion: false
level_of_theory: r2SCAN
kwargs:
model: "pet-mad-s"
version: "1.5.0"
dispersion_kwargs:
xc: r2scan
pet-oam-xl¶
pet-oam-xl:
module: upet.calculator
class_name: UPETCalculator
device: "cpu"
trained_on_dispersion: False
level_of_theory: PBE
kwargs:
model: "pet-oam-xl"
version: "1.0.0"
UMA (FairChem)¶
uma-s-1p1-omat¶
uma-s-1p1-omat:
module: fairchem.core
class_name: FAIRChemCalculator
device: "cpu"
level_of_theory: PBE
trained_on_dispersion: false
kwargs:
model_name: "uma-s-1p1"
task_name: "omat"
uma-m-1p1-omat¶
uma-m-1p1-omat:
module: fairchem.core
class_name: FAIRChemCalculator
device: "cpu"
level_of_theory: PBE
trained_on_dispersion: false
kwargs:
model_name: "uma-m-1p1"
task_name: "omat"
uma-s-1p1-omol¶
uma-s-1p1-omol:
module: fairchem.core
class_name: FAIRChemCalculator
device: "cpu"
trained_on_dispersion: true
level_of_theory: ωB97M-V/def2-TZVPD
kwargs:
model_name: "uma-s-1p1"
task_name: "omol"
uma-s-1p2-omol¶
uma-s-1p2-omol:
module: fairchem.core
class_name: FAIRChemCalculator
device: "cpu"
trained_on_dispersion: true
level_of_theory: ωB97M-V/def2-TZVPD
kwargs:
model_name: "uma-s-1p2"
task_name: "omol"
uma-m-1p1-omol¶
uma-m-1p1-omol:
module: fairchem.core
class_name: FAIRChemCalculator
device: "cpu"
trained_on_dispersion: true
level_of_theory: ωB97M-V/def2-TZVPD
kwargs:
model_name: "uma-m-1p1"
task_name: "omol"
MatterSim¶
mattersim-5M¶
mattersim-5M:
module: mattersim.forcefield
class_name: MatterSimCalculator
device: "cpu"
trained_on_dispersion: false
level_of_theory: PBE
kwargs:
load_path: "mattersim-v1.0.0-5m"
mattersim-1M¶
mattersim-1M:
module: mattersim.forcefield
class_name: MatterSimCalculator
device: "cpu"
trained_on_dispersion: false
level_of_theory: PBE
kwargs:
load_path: "mattersim-v1.0.0-1m"
GRACE¶
GRACE-2L-OMAT¶
grace-2l-omat:
module: tensorpotential.calculator
class_name: grace_fm
device: "cpu"
trained_on_dispersion: false
kwargs:
model: GRACE-2L-OMAT
GRACE-2L-SMAX-OMAT-medium¶
grace-2l-smax-omat-medium:
module: tensorpotential.calculator
class_name: grace_fm
device: "cpu"
trained_on_dispersion: false
kwargs:
model: GRACE-2L-SMAX-OMAT-medium
SevenNet¶
sevennet-omni-mpa¶
sevennet-omni-mpa:
module: sevenn.sevennet_calculator
class_name: SevenNetCalculator
device: "cpu"
trained_on_dispersion: false
level_of_theory: PBE
kwargs:
model: "7net-omni"
modal: "mpa"