Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

add projection layer to roberta encoder #1451

Open
wants to merge 2 commits into
base: main
Choose a base branch
from
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 12 additions & 0 deletions torchtext/models/roberta/model.py
Original file line number Diff line number Diff line change
Expand Up @@ -10,6 +10,7 @@

from .modules import (
TransformerEncoder,
ProjectionLayer,
)
import logging
logger = logging.getLogger(__name__)
Expand All @@ -25,6 +26,8 @@ class RobertaEncoderConf:
num_attention_heads: int = 12
num_encoder_layers: int = 12
dropout: float = 0.1
projection_dim: Optional[int] = None
projection_dropout: Optional[float] = None
scaling: Optional[float] = None
normalize_before: bool = False

Expand All @@ -40,6 +43,8 @@ def __init__(
num_attention_heads: int,
num_encoder_layers: int,
dropout: float = 0.1,
projection_dim: Optional[int] = None,
projection_dropout: Optional[float] = None,
scaling: Optional[float] = None,
normalize_before: bool = False,
):
Expand All @@ -62,6 +67,10 @@ def __init__(
return_all_layers=False,
)

self.project = nn.Identity()
if projection_dim is not None:
self.project = ProjectionLayer(embed_dim=embedding_dim, projection_dim=projection_dim, dropout=projection_dropout)

@classmethod
def from_config(cls, config: RobertaEncoderConf):
return cls(**asdict(config))
Expand All @@ -73,6 +82,9 @@ def forward(self, tokens: Tensor, mask: Optional[Tensor] = None) -> Tensor:
output = output.transpose(1, 0)
if mask is not None:
output = output[mask.to(torch.bool), :]

output = self.project(output)

return output


Expand Down
21 changes: 21 additions & 0 deletions torchtext/models/roberta/modules.py
Original file line number Diff line number Diff line change
Expand Up @@ -31,6 +31,27 @@ def _make_positions(self, tensor, pad_index: int):
return torch.cumsum(masked, dim=1) * masked + pad_index


class ProjectionLayer(Module):
def __init__(self,
embed_dim: int,
projection_dim: int,
dropout: Optional[float] = None) -> None:
super().__init__()

self.projection_layer = nn.Linear(embed_dim, projection_dim)
self.norm_layer = nn.LayerNorm(projection_dim)
if dropout is not None:
self.dropout_layer = nn.Dropout(dropout)
else:
self.dropout_layer = nn.Identity()

def forward(self, x: torch.Tensor) -> torch.Tensor:
x = self.projection_layer(x)
x = self.norm_layer(x)
x = self.dropout_layer(x)
return x


class ResidualMLP(Module):
def __init__(
self,
Expand Down