SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

Support return_tensors in audio chat templates (#34601)

Browse Source 
* add audio chat templates

* update

* update

* nit

* green ci

* we dont care about the order anymore

* clean up after rebase

* overriden tests rename

* rename shieldgemma also

* one more rename

* require_read_token

* removde images/videos

* retrigger CI flaky
This commit is contained in:
Raushan Turganbay
2025-03-25 11:08:47 +01:00
committed by GitHub
parent 19085c28da
commit 0f733110a6
11 changed files with 403 additions and 160 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
tests/models/wav2vec2/test_processor_wav2vec2.py
View File

@@ -18,8 +18,6 @@ import shutil
import tempfile
import unittest
import numpy as np
from transformers.models.wav2vec2 import Wav2Vec2CTCTokenizer, Wav2Vec2FeatureExtractor, Wav2Vec2Processor
from transformers.models.wav2vec2.tokenization_wav2vec2 import VOCAB_FILES_NAMES
from transformers.utils import FEATURE_EXTRACTOR_NAME
@@ -30,6 +28,8 @@ from .test_feature_extraction_wav2vec2 import floats_list
class Wav2Vec2ProcessorTest(ProcessorTesterMixin, unittest.TestCase):
processor_class = Wav2Vec2Processor
audio_input_name = "input_values"
text_input_name = "labels"
def setUp(self):
vocab = "<pad> <s> </s> <unk> | E T A O N I H S R D L U M W C F G Y P B V K ' X J Q Z".split(" ")
@@ -132,22 +132,6 @@ class Wav2Vec2ProcessorTest(ProcessorTesterMixin, unittest.TestCase):
for key in encoded_tok.keys():
self.assertListEqual(encoded_tok[key], encoded_processor[key])
def test_padding_argument_not_ignored(self):
# padding, or any other overlap arg between audio extractor and tokenizer
# should be passed to both text and audio and not ignored
feature_extractor = self.get_feature_extractor()
tokenizer = self.get_tokenizer()
processor = Wav2Vec2Processor(tokenizer=tokenizer, feature_extractor=feature_extractor)
batch_duration_in_seconds = [1, 3, 2, 6]
input_features = [np.random.random(16_000 * s) for s in batch_duration_in_seconds]
# padding = True should not raise an error and will if the audio processor popped its value to None
_ = processor(
input_features, padding=True, sampling_rate=processor.feature_extractor.sampling_rate, return_tensors="pt"
)
def test_tokenizer_decode(self):
feature_extractor = self.get_feature_extractor()
tokenizer = self.get_tokenizer()