Skip to content

Refactor STT runtime around model-owned adapters #173

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
ericcurtin merged 7 commits into from
Mar 19, 2026
Merged

Refactor STT runtime around model-owned adapters #173

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
e0f100d
Extract STTExecutor into stt runtime module
LxYuan0420 Mar 17, 2026
c462447
Add model-owned STT runtime adapters
LxYuan0420 Mar 18, 2026
736152d
Type STT runtime adapter creation
LxYuan0420 Mar 18, 2026
f2c76d2
Remove legacy STT transcribe module
LxYuan0420 Mar 18, 2026
b3f5843
Normalize STT model_type declarations
LxYuan0420 Mar 18, 2026
757b677
Fix lint import ordering in STT tests
LxYuan0420 Mar 18, 2026
990ab57
Apply ruff formatting to STT runtime files
LxYuan0420 Mar 18, 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
3 changes: 2 additions & 1 deletion tests/test_qwen3_asr.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12,6 +12,7 @@
import pytest

from vllm_metal.stt.config import is_stt_model
from vllm_metal.stt.loader import load_model
from vllm_metal.stt.qwen3_asr.config import (
Qwen3ASRAudioConfig,
Qwen3ASRConfig,
Expand All @@ -23,7 +24,7 @@
Qwen3Attention,
Qwen3LM,
)
from vllm_metal.stt.transcribe import Qwen3ASRTranscriber, load_model
from vllm_metal.stt.qwen3_asr.transcriber import Qwen3ASRTranscriber

# ===========================================================================
# Configuration
Expand Down
18 changes: 9 additions & 9 deletions tests/test_transcribe.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,11 +10,9 @@
import mlx.core as mx
import pytest

from vllm_metal.stt.transcribe import (
TranscriptionResult,
WhisperTranscriber,
load_model,
)
from vllm_metal.stt.loader import load_model
from vllm_metal.stt.protocol import TranscriptionResult
from vllm_metal.stt.whisper import WhisperTranscriber
from vllm_metal.stt.whisper.transcriber import MAX_PROMPT_TOKENS

# ===========================================================================
Expand Down Expand Up @@ -370,11 +368,12 @@ def test_transcribe_silence(self) -> None:
import mlx.core as mx

from vllm_metal.stt.audio import SAMPLE_RATE
from vllm_metal.stt.transcribe import load_model, transcribe
from vllm_metal.stt.loader import load_model
from vllm_metal.stt.whisper import WhisperTranscriber

model = load_model("openai/whisper-tiny")
audio = mx.zeros(SAMPLE_RATE * 3) # 3s silence
result = transcribe(model, audio)
result = WhisperTranscriber(model).transcribe(audio)
# Whisper may hallucinate on silence; just check it doesn't crash
assert isinstance(result.text, str)

Expand All @@ -383,9 +382,10 @@ def test_transcribe_with_timestamps(self) -> None:
import mlx.core as mx

from vllm_metal.stt.audio import SAMPLE_RATE
from vllm_metal.stt.transcribe import load_model, transcribe
from vllm_metal.stt.loader import load_model
from vllm_metal.stt.whisper import WhisperTranscriber

model = load_model("openai/whisper-tiny")
audio = mx.zeros(SAMPLE_RATE * 3)
result = transcribe(model, audio, with_timestamps=True)
result = WhisperTranscriber(model).transcribe(audio, with_timestamps=True)
assert result.duration == pytest.approx(3.0)
Loading
Loading