[Refactor] Remove multiple dead codes (#50285)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
This commit is contained in:
@@ -77,24 +77,6 @@ class BartScaledWordEmbedding(VocabParallelEmbedding):
|
||||
return super().forward(input_ids) * self.embed_scale
|
||||
|
||||
|
||||
class BartParallelLMHead(ParallelLMHead):
|
||||
"""
|
||||
This module overrides ParallelLMHead's
|
||||
forward by dividing by embeddings scale,
|
||||
yielding effectively the inverse of
|
||||
BartScaledWordEmbedding
|
||||
"""
|
||||
|
||||
def __init__(
|
||||
self, num_embeddings: int, embedding_dim: int, embed_scale: float = 1.0
|
||||
):
|
||||
super().__init__(num_embeddings, embedding_dim)
|
||||
self.embed_scale = embed_scale
|
||||
|
||||
def forward(self, input_ids: torch.Tensor) -> torch.Tensor:
|
||||
return super().forward(input_ids) / self.embed_scale
|
||||
|
||||
|
||||
class BartDecoderLayer(nn.Module):
|
||||
def __init__(
|
||||
self,
|
||||
|
||||
Reference in New Issue
Block a user