A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 7, 20:15 UTC
[Bugfix] Fix failing FunASR processor test (#35111)
Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn>
Isotr0py committed
Feb 24, 2026 at 20:13 UTC
d12d20140949ada88546d3f13ae68d4cea070a7e
2 files changed
+2
-4
tests/models/multimodal/processing/test_common.py
-2
@@ -460,8 +460,6 @@ def test_processing_correctness(
460
num_batches: int,
461
simplify_rate: float,
462
):
463
- if model_id == "allendou/Fun-ASR-Nano-2512-vllm":
464
- pytest.skip("Cached audio `input_features` not matched. Fix later.")
463
if model_id == "google/gemma-3n-E2B-it":
464
pytest.skip("Fix later")
465
if model_id == "OpenGVLab/InternVL2-2B":
vllm/transformers_utils/processors/funasr_processor.py
+2
-2
@@ -361,11 +361,11 @@ class FunASRFeatureExtractor(SequenceFeatureExtractor):
361
362
input_features = padded_inputs.get("input_features").transpose(2, 0, 1)
363
364
- self.frontend = WavFrontend(**self.frontend_conf)
364
+ frontend = WavFrontend(**self.frontend_conf, dither=self.dither)
365
input_features, speech_lengths = self.extract_fbank(
366
input_features[0],
367
data_type=kwargs.get("data_type", "sound"),
368
- frontend=self.frontend,
368
+ frontend=frontend,
369
is_final=True,
370
)
371
olens = 1 + (speech_lengths - 3 + 2 * 1) // 2