Skip to content

fix(transformer): enforce max_len in make_transformer_src_tgt_masks #3086

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

Open
FinalSunFlower wants to merge 1 commit into
base: develop
Choose a base branch
from
+25 −1
Open

fix(transformer): enforce max_len in make_transformer_src_tgt_masks #3086

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
2 changes: 1 addition & 1 deletion speechbrain/lobes/models/transformer/TransformerASR.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -146,7 +146,7 @@ def make_transformer_src_tgt_masks(
# mask out audio beyond the length of audio for each batch
if wav_len is not None:
abs_len = torch.round(wav_len * src.shape[1])
src_key_padding_mask = ~length_to_mask(abs_len).bool()
src_key_padding_mask = ~length_to_mask(abs_len, max_len=src.shape[1]).bool()

# mask out the source
src_mask = make_transformer_src_mask(
Expand Down
24 changes: 24 additions & 0 deletions tests/unittests/test_transformer_src_tgt_masks.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,6 +2,30 @@
import torch.nn


def test_src_key_padding_mask_matches_padded_width(device):
from speechbrain.lobes.models.transformer.TransformerASR import (
make_transformer_src_tgt_masks,
)

src = torch.randn(3, 100, 80, device=device)
wav_len = torch.tensor([0.50, 0.75, 0.90], device=device)

src_key_padding_mask, _, _, _ = make_transformer_src_tgt_masks(src, wav_len=wav_len)

assert src_key_padding_mask.shape == (3, src.shape[1])

attention = torch.nn.MultiheadAttention(
embed_dim=src.shape[-1], num_heads=1, batch_first=True
).to(device)
output, _ = attention(
src,
src,
src,
key_padding_mask=src_key_padding_mask,
)
assert output.shape == src.shape


def test_make_transformer_src_tgt_masks(device):
from numpy import inf

Expand Down