models/equiformerv2/model_configs.py

model_config_orig = {
    "use_pbc": True, 
    "regress_forces": True,
    "otf_graph": True,  
    "max_neighbors": 20,  
    "max_radius": 8.0, # 12.0, #NOTE changed from 12 to 8  
    "max_num_elements": 23,  
    "num_layers": 8,
    "sphere_channels": 128,
    "attn_hidden_channels": 64,
    "num_heads": 8,
    "attn_alpha_channels": 64,
    "attn_value_channels": 16,
    "ffn_hidden_channels": 128,
    "norm_type": "layer_norm_sh",
    "lmax_list": [4],
    "mmax_list": [2],
    "grid_resolution": 18,
    "num_sphere_samples": 128,
    "edge_channels": 128,
    "use_atom_edge_embedding": True,
    "share_atom_edge_embedding": False,
    "distance_function": "gaussian",
    "num_distance_basis": 512,
    "attn_activation": "silu",
    "use_s2_act_attn": False,
    "use_attn_renorm": True,
    "ffn_activation": "silu",
    "use_gate_act": False,
    "use_grid_mlp": True,
    "use_sep_s2_act": True,
    "alpha_drop": 0.1,
    "drop_path_rate": 0.1,
    "proj_drop": 0.0,
    "weight_init": "uniform",
}


model_config_small = { # ca 4.8M params
    "use_pbc": True, 
    "regress_forces": True,
    "otf_graph": True,  
    "max_neighbors": 20,  
    "max_radius": 8.0,  
    "max_num_elements": 23,  
    "num_layers": 5, #8,
    "sphere_channels": 64, #128,
    "attn_hidden_channels": 32, #64,
    "num_heads": 4, # 8
    "attn_alpha_channels": 32, #64,
    "attn_value_channels": 16,
    "ffn_hidden_channels": 64, #128,
    "norm_type": "layer_norm_sh",
    "lmax_list": [4],
    "mmax_list": [2],
    "grid_resolution": 18,
    "num_sphere_samples": 128,
    "edge_channels": 64, #128,
    "use_atom_edge_embedding": True,
    "share_atom_edge_embedding": False,
    "distance_function": "gaussian",
    "num_distance_basis": 512,
    "attn_activation": "silu",
    "use_s2_act_attn": False,
    "use_attn_renorm": True,
    "ffn_activation": "silu",
    "use_gate_act": False,
    "use_grid_mlp": True,
    "use_sep_s2_act": True,
    "alpha_drop": 0.1,
    "drop_path_rate": 0.1,
    "proj_drop": 0.0,
    "weight_init": "uniform",
}