lgatr.layers.mlp.config.MLPConfig

class lgatr.layers.mlp.config.MLPConfig(mv_channels=None, s_channels=None, dropout_prob=None, activation='gelu', increase_hidden_channels=2, num_hidden_layers=1)[source]

Bases: object

Geometric MLP configuration.

Parameters:

  • activation ({"relu", "sigmoid", "gelu"}) – Which (gated) activation function to use.

  • increase_hidden_channels (int) – Factor by which to increase the number of hidden channels (both multivectors and scalars). Vanilla transformers use 4, we use 2 for backward compatibility.

  • num_hidden_layers (int) – Number of hidden layers to create.

Parameters auto-set by LGATr:

  • mv_channels (int) – Number of input multivector channels.

  • s_channels (int) – Number of input scalar channels.

  • dropout_prob (float or None) – Dropout probability

activation: str = 'gelu'
classmethod cast(config)[source]

Casts an object as MLPConfig.

Return type:

MLPConfig

dropout_prob: Optional[float] = None
increase_hidden_channels: int = 2
mv_channels: Optional[int] = None
num_hidden_layers: int = 1
s_channels: Optional[int] = None