nb_heads = nb_heads,
causal = True, attention_dropout = dropout
),
- nn.Linear(in_features = dim_model, out_features = dim_model),
),
Residual(
nn.LayerNorm(dim_model),
self.readout = nn.Linear(in_features = dim_model, out_features = vocabulary_size)
def forward(self, x):
+ x = F.pad(x, (1, 0))
x = self.embedding(x)
x = self.trunk(x)
x = self.readout(x)
- return x
+ return x[:, :-1]
######################################################################