Skip to content

[Bugfix][Model][Rust Frontend] Qwen3_ASR: Handle ragged Qwen audio batches from Rust frontend #60979

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
Isotr0py merged 3 commits into from
Oct 13, 2026
Merged

[Bugfix][Model][Rust Frontend] Qwen3_ASR: Handle ragged Qwen audio batches from Rust frontend #60979

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
32e87ef
[Bugfix][Model][Rust Frontend] Qwen3_ASR: Handle ragged Qwen audio ba…
noone-zhhh Oct 10, 2026
8df7c1b
[Test] Remove unit test
noone-zhhh Oct 12, 2026
103f72e
Merge branch 'main' into bugfix/qwen_asr_audio_input
Isotr0py Oct 13, 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
12 changes: 7 additions & 5 deletions vllm/model_executor/models/qwen2_5_omni_thinker.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -132,11 +132,11 @@ def _get_feature_tensor(feature: MultiModalFeatureSpec, key: str) -> torch.Tenso


def unpad_and_flat_audio_features(
input_audio_features: torch.Tensor,
input_audio_features: torch.Tensor | Sequence[torch.Tensor],
audio_feature_lengths: torch.Tensor,
) -> torch.Tensor:
"""Unpad and flatten batched audio features."""
if len(audio_feature_lengths) != input_audio_features.size(0):
"""Unpad and flatten stacked or ragged per-audio features."""
if len(audio_feature_lengths) != len(input_audio_features):
raise ValueError(
"Length of audio_feature_lengths must match "
"the batch size of input_audio_features."
Expand Down Expand Up @@ -920,8 +920,10 @@ def _parse_and_validate_audio_input(

# inputs features from rust frontend is batched and padded
# with shape [batch_size, n_mels, padded_seq_len], different
# from python's shape [n_mels, batch_size * seq_len]
if (
# from python's shape [n_mels, batch_size * seq_len].
# Requests with different padded lengths are batched as a
# list of [n_mels, padded_seq_len] tensors.
if isinstance(input_audio_features, (list, tuple)) or (
isinstance(input_audio_features, torch.Tensor)
and input_audio_features.dim() == 3
):
Expand Down
6 changes: 4 additions & 2 deletions vllm/model_executor/models/qwen3_asr.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -478,8 +478,10 @@ def _parse_and_validate_audio_input(

# inputs features from rust frontend is batched and padded
# with shape [batch_size, n_mels, padded_seq_len], different
# from python's shape [n_mels, batch_size * seq_len]
if (
# from python's shape [n_mels, batch_size * seq_len].
# Requests with different padded lengths are batched as a
# list of [n_mels, padded_seq_len] tensors.
if isinstance(input_audio_features, (list, tuple)) or (
isinstance(input_audio_features, torch.Tensor)
and input_audio_features.dim() == 3
):
Expand Down
Loading