Skip to content

Fixing bug in Voxtral when merging text and audio embeddings #40671

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
merged 5 commits into from
Sep 4, 2025
Merged

Fixing bug in Voxtral when merging text and audio embeddings #40671

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
f9bbbbc
Fixing bug when replacing text-audio token placeholders with audio em…
rcogill Sep 3, 2025
47fa46e
apply changes
eustlb Sep 4, 2025
bd0751c
Merge branch 'main' into dev-voxtral-embedding-bugfix
eustlb Sep 4, 2025
eb17a17
audio token replacement does not make sense when input ids not provided
eustlb Sep 4, 2025
988cb9c
Merge branch 'main' into dev-voxtral-embedding-bugfix
eustlb Sep 4, 2025
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
8 changes: 5 additions & 3 deletions src/transformers/models/voxtral/modeling_voxtral.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -504,12 +504,14 @@ def forward(
if inputs_embeds is None:
inputs_embeds = self.get_input_embeddings()(input_ids)

if input_features is not None:
if input_features is not None and input_ids is not None:
audio_embeds = self.get_audio_embeds(input_features)

# replace text-audio token placeholders with audio embeddings
audio_token_mask = input_ids == self.config.audio_token_id
inputs_embeds[audio_token_mask] = audio_embeds
audio_token_mask = (input_ids == self.config.audio_token_id).unsqueeze(-1)
inputs_embeds = inputs_embeds.masked_scatter(
audio_token_mask.to(inputs_embeds.device), audio_embeds.to(inputs_embeds.device)
)

outputs: BaseModelOutputWithPast = self.language_model(
attention_mask=attention_mask,
Expand Down
8 changes: 5 additions & 3 deletions src/transformers/models/voxtral/modular_voxtral.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -239,12 +239,14 @@ def forward(
if inputs_embeds is None:
inputs_embeds = self.get_input_embeddings()(input_ids)

if input_features is not None:
if input_features is not None and input_ids is not None:
audio_embeds = self.get_audio_embeds(input_features)

# replace text-audio token placeholders with audio embeddings
audio_token_mask = input_ids == self.config.audio_token_id
inputs_embeds[audio_token_mask] = audio_embeds
audio_token_mask = (input_ids == self.config.audio_token_id).unsqueeze(-1)
inputs_embeds = inputs_embeds.masked_scatter(
audio_token_mask.to(inputs_embeds.device), audio_embeds.to(inputs_embeds.device)
)

outputs: BaseModelOutputWithPast = self.language_model(
attention_mask=attention_mask,
Expand Down