def __init__(self, config): super(SpanAttention, self).__init__() # create modules self.attention = SpanWordAttention(config) self.output = BertSelfOutput(config) # initialize weights self.init_weights()
def __init__(self, config, add_conditional_layernorm=True): super().__init__(config) self.self = MyBertSelfAttention10(config) self.add_conditional_layernorm = add_conditional_layernorm if add_conditional_layernorm: self.output = MyBertSelfOutput10(config) else: self.output = BertSelfOutput(config) self.pruned_heads = set()
def __init__(self, config: BertConfig, bias=False): super().__init__() self.config = config self.head_num = config.num_attention_heads self.head_dim = config.hidden_size // self.head_num self.all_dim = self.head_num * self.head_dim self.query = nn.Linear(config.hidden_size, self.all_dim, bias=bias) self.kv = nn.Linear(config.hidden_size, self.all_dim, bias=bias) self.output = BertSelfOutput(config) self.dropout = nn.Dropout(config.attention_probs_dropout_prob) self.scale = 1 / (self.head_dim**0.5)
def __init__(self, config): super(EntityAwareAttention, self).__init__() self.self = EntityAwareSelfAttention(config) self.output = BertSelfOutput(config)
def __init__(self, config): super(BertAttention, self).__init__() self.self = BertSelfAttention(config) self.output = BertSelfOutput(config) self.pruned_heads = set()
def __init__(self, config): super().__init__() self.att = BertSelfAttention(config) self.output = BertSelfOutput(config)
def __init__(self, config): super().__init__() self.self = BertScanAttentionHeads(config) self.output = BertSelfOutput(config)
def __init__(self, config): super(CaptionBertAttention, self).__init__(config) self.self = CaptionBertSelfAttention(config) self.output = BertSelfOutput(config)
def __init__(self, config): super(BertAttention_attention, self).__init__() self.self = BertSelfAttention_attention(config) self.output = BertSelfOutput(config)