Skip to content

fix(attention): fix wrong broadcast #52

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.

Sign up for GitHub

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

Merged
shenxiangzhuang merged 1 commit into from
Nov 14, 2025
Merged

fix(attention): fix wrong broadcast #52

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
3 changes: 1 addition & 2 deletions toynlp/bert/model.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -60,8 +60,7 @@ def forward(
attention_weight = q @ k.transpose(-2, -1) / (head_dim**0.5)
# pad mask
if mask is not None:
# TODO: -inf?
attention_weight = attention_weight.masked_fill(mask == 0, float("-10000"))
attention_weight = attention_weight.masked_fill(mask == 0, float("-inf"))

attention_score = torch.nn.functional.softmax(attention_weight, dim=-1)
attention_score = self.dropout(attention_score)
Expand Down
6 changes: 3 additions & 3 deletions toynlp/transformer/model.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -54,7 +54,7 @@ def forward(
# pad mask
if mask is not None:
# TODO: -inf?
attention_weight = attention_weight.masked_fill(mask == 0, float("-10000"))
attention_weight = attention_weight.masked_fill(mask == 0, float("-inf"))

attention_score = torch.nn.functional.softmax(attention_weight, dim=-1)

Expand Down Expand Up @@ -253,8 +253,8 @@ def _get_source_mask(self, source_token_ids: torch.Tensor) -> torch.Tensor:
return (source_token_ids != self.padding_idx).unsqueeze(1).unsqueeze(2)

def _get_target_mask(self, target_token_ids: torch.Tensor) -> torch.Tensor:
# shape: (batch_size, 1, target_seq_length, 1)
pad_mask = (target_token_ids != self.padding_idx).unsqueeze(1).unsqueeze(3)
# shape: (batch_size, 1, 1, target_seq_length)
pad_mask = (target_token_ids != self.padding_idx).unsqueeze(1).unsqueeze(2)
target_seq_length = target_token_ids.size(1)
# shape: (batch_size, 1, target_seq_length, target_seq_length)
causal_mask = torch.tril(torch.ones((target_seq_length, target_seq_length), device=self.device)).bool()
Expand Down
Loading