def __init__(
self,
hidden_size,
num_intents,
num_slots,
dropout=0.0,
use_transformer_pretrained=True,
**kwargs,
):
super().__init__(**kwargs)
self.dropout = nn.Dropout(dropout)
self.slot_mlp = MultiLayerPerceptron(
hidden_size,
num_classes=num_slots,
device=self._device,
num_layers=2,
activation='relu',
log_softmax=False,
)
self.intent_mlp = MultiLayerPerceptron(
hidden_size,
num_classes=num_intents,
device=self._device,
num_layers=2,
activation='relu',
log_softmax=False,
)
if use_transformer_pretrained:
self.apply(
lambda module: transformer_weights_init(module, xavier=False))
def __init__(
self,
hidden_size,
num_classes,
activation='relu',
log_softmax=True,
dropout=0.0,
use_transformer_pretrained=True,
):
super().__init__()
if activation not in ACT2FN:
raise ValueError(f'activation "{activation}" not found')
self.dense = nn.Linear(hidden_size, hidden_size)
self.act = ACT2FN[activation]
self.norm = nn.LayerNorm(hidden_size, eps=1e-12)
self.mlp = MultiLayerPerceptron(
hidden_size,
num_classes,
self._device,
num_layers=1,
activation=activation,
log_softmax=log_softmax,
)
self.dropout = nn.Dropout(dropout)
if use_transformer_pretrained:
self.apply(
lambda module: transformer_weights_init(module, xavier=False))
self.to(self._device)
def __init__(
self,
hidden_size,
num_classes,
name=None,
num_layers=2,
activation='relu',
log_softmax=True,
dropout=0.0,
use_transformer_pretrained=True,
):
super().__init__()
self.name = name
self.mlp = MultiLayerPerceptron(
hidden_size,
num_classes,
self._device,
num_layers,
activation,
log_softmax,
)
self.dropout = nn.Dropout(dropout)
if use_transformer_pretrained:
self.apply(
lambda module: transformer_weights_init(module, xavier=False))
def __init__(self,
hidden_size,
num_classes,
activation=nn.functional.relu,
log_softmax=True,
dropout=0.0,
use_transformer_pretrained=True):
super().__init__()
self.dense = nn.Linear(hidden_size, hidden_size)
self.act = activation
self.norm = nn.LayerNorm(hidden_size, eps=1e-12)
self.mlp = MultiLayerPerceptron(hidden_size,
num_classes,
self._device,
num_layers=1,
activation=activation,
log_softmax=log_softmax)
self.dropout = nn.Dropout(dropout)
if use_transformer_pretrained:
self.apply(
lambda module: transformer_weights_init(module, xavier=False))
self.to(self._device)