Skip to content

Fix inconsistent return type in TextEmbedder._get_sentence_embedding #67

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

Closed
Copilot wants to merge 3 commits into from
Closed

Fix inconsistent return type in TextEmbedder._get_sentence_embedding #67

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
3ef803e
Initial plan
Copilot Jan 27, 2026
ed56ec5
Fix: return consistent dict structure from _get_sentence_embedding fo…
Copilot Jan 27, 2026
51c6d6b
Update return type annotation for _get_sentence_embedding to reflect …
Copilot Jan 27, 2026
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
24 changes: 15 additions & 9 deletions torchTextClassifiers/model/components/text_embedder.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -198,15 +198,15 @@ def _get_sentence_embedding(
token_embeddings: torch.Tensor,
attention_mask: torch.Tensor,
return_label_attention_matrix: bool = False,
) -> torch.Tensor:
) -> dict[str, Optional[torch.Tensor]]:
"""
Compute sentence embedding from embedded tokens - "remove" second dimension.

Args (output from dataset collate_fn):
token_embeddings (torch.Tensor[Long]), shape (batch_size, seq_len, embedding_dim): Tokenized + padded text
attention_mask (torch.Tensor[Long]), shape (batch_size, seq_len): Attention mask indicating non-pad tokens
Returns:
torch.Tensor: Sentence embeddings, shape (batch_size, embedding_dim)
dict: Dictionary with keys 'sentence_embedding' (torch.Tensor) and 'label_attention_matrix' (Optional[torch.Tensor])
"""

# average over non-pad token embeddings
Expand All @@ -217,14 +217,20 @@ def _get_sentence_embedding(
if self.attention_config is not None:
if self.attention_config.aggregation_method is not None: # default is "mean"
if self.attention_config.aggregation_method == "first":
return token_embeddings[:, 0, :]
return {
"sentence_embedding": token_embeddings[:, 0, :],
"label_attention_matrix": None,
}
elif self.attention_config.aggregation_method == "last":
lengths = attention_mask.sum(dim=1).clamp(min=1) # last non-pad token index + 1
return token_embeddings[
torch.arange(token_embeddings.size(0)),
lengths - 1,
:,
]
lengths = attention_mask.sum(dim=1).clamp(min=1).long() # last non-pad token index + 1
return {
"sentence_embedding": token_embeddings[
torch.arange(token_embeddings.size(0)),
lengths - 1,
:,
],
"label_attention_matrix": None,
}
else:
if self.attention_config.aggregation_method != "mean":
raise ValueError(
Expand Down