_{grace-off — ML potentials for liquids}

Models

The GRACE-OFF models are available in the models folder.
The 1-layer models are stored in models/1l, and the 2-layer models are stored in models/2l.

To run an MD simulation using the ASE calculator, you can use the following code snippet:

from tensorpotential.calculator import TPCalculator

mol.calc = TPCalculator(
    model=f"{model_path}",
    device="cuda",
    float_dtype=default_dtype,
)

Different model sizes for the 1LAYER and 2LAYER models

# Layers	Model	# Parameters	l_max	max_order	n_rad_max	n_mlp_dens
1	small	58 072	3	3	20	10
1	medium	512 128	4	4	32	12
1	large	1 026 592	4	4	48	16
2	small	235 976	[3, 2]	3	[20, 32]	8
2	medium	901 300	[3, 3]	4	[32, 42]	10
2	large	1 626 848	[4, 3]	4	[32, 48]	12

These are the parameters (taken from here)

• lmax, maximum per-layer l-character for constructing product functions
• max_order, maximum product order for B-functions
• n_rad_max, per-layer max number of radial functions
• n_mlp_dens, number of non-linear readout densities