def __init__(self, args, dictionary, embed_tokens): self.args = args super().__init__( TransformerConfig.from_namespace(args), dictionary, embed_tokens, )
def build_self_attention( self, embed_dim, args, add_bias_kv=False, add_zero_attn=False ): return super().build_self_attention( embed_dim, TransformerConfig.from_namespace(args), add_bias_kv=add_bias_kv, add_zero_attn=add_zero_attn, )
def __init__( self, args, no_encoder_attn=False, add_bias_kv=False, add_zero_attn=False ): super().__init__( TransformerConfig.from_namespace(args), no_encoder_attn=no_encoder_attn, add_bias_kv=add_bias_kv, add_zero_attn=add_zero_attn, ) self.args = args
def __init__( self, args, positional_embedding: Optional[RelativePositionalEmbedding] = None ): super().__init__( TransformerConfig.from_namespace(args), positional_embedding=positional_embedding, ) self.args = args
def __init__( self, args, dictionary, embed_tokens, no_encoder_attn=False, output_projection=None, ): self.args = args super().__init__( TransformerConfig.from_namespace(args), dictionary, embed_tokens, no_encoder_attn=no_encoder_attn, output_projection=output_projection, )
def __init__( self, args, no_encoder_attn=False, add_bias_kv=False, add_zero_attn=False, positional_embedding: Optional[RelativePositionalEmbedding] = None, ): super().__init__( TransformerConfig.from_namespace(args), no_encoder_attn=no_encoder_attn, add_bias_kv=add_bias_kv, add_zero_attn=add_zero_attn, positional_embedding=positional_embedding, ) self.args = args
def build_encoder_attention(self, embed_dim, args): return super().build_encoder_attention( embed_dim, TransformerConfig.from_namespace(args), )
def __init__(self, args): super().__init__(TransformerConfig.from_namespace(args)) self.args = args
def build_encoder_layer(self, args): return super().build_encoder_layer( TransformerConfig.from_namespace(args), )
def build_decoder_layer(self, args, no_encoder_attn=False): return super().build_decoder_layer( TransformerConfig.from_namespace(args), no_encoder_attn=no_encoder_attn)
def build_output_projection(self, args, dictionary, embed_tokens): super().build_output_projection(TransformerConfig.from_namespace(args), dictionary, embed_tokens)
def build_self_attention(self, embed_dim, args, positional_embedding=None): return super().build_self_attention( embed_dim, TransformerConfig.from_namespace(args), positional_embedding=positional_embedding, )