chendl/compositional_test

1# Copyright 2021 The HuggingFace Team. All rights reserved.2#3# Licensed under the Apache License, Version 2.0 (the "License");4# you may not use this file except in compliance with the License.5# You may obtain a copy of the License at6#7#     http://www.apache.org/licenses/LICENSE-2.08#9# Unless required by applicable law or agreed to in writing, software10# distributed under the License is distributed on an "AS IS" BASIS,11# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.12# See the License for the specific language governing permissions and13# limitations under the License.14 15import unittest16 17import numpy as np18 19from transformers import MODEL_FOR_AUDIO_CLASSIFICATION_MAPPING20from transformers.pipelines import AudioClassificationPipeline, pipeline21from transformers.testing_utils import (22    is_pipeline_test,23    nested_simplify,24    require_tf,25    require_torch,26    require_torchaudio,27    slow,28)29 30from .test_pipelines_common import ANY31 32 33@is_pipeline_test34@require_torch35class AudioClassificationPipelineTests(unittest.TestCase):36    model_mapping = MODEL_FOR_AUDIO_CLASSIFICATION_MAPPING37 38    def get_test_pipeline(self, model, tokenizer, processor):39        audio_classifier = AudioClassificationPipeline(model=model, feature_extractor=processor)40 41        # test with a raw waveform42        audio = np.zeros((34000,))43        audio2 = np.zeros((14000,))44        return audio_classifier, [audio2, audio]45 46    def run_pipeline_test(self, audio_classifier, examples):47        audio2, audio = examples48        output = audio_classifier(audio)49        # by default a model is initialized with num_labels=250        self.assertEqual(51            output,52            [53                {"score": ANY(float), "label": ANY(str)},54                {"score": ANY(float), "label": ANY(str)},55            ],56        )57        output = audio_classifier(audio, top_k=1)58        self.assertEqual(59            output,60            [61                {"score": ANY(float), "label": ANY(str)},62            ],63        )64 65        self.run_torchaudio(audio_classifier)66 67    @require_torchaudio68    def run_torchaudio(self, audio_classifier):69        import datasets70 71        # test with a local file72        dataset = datasets.load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation")73        audio = dataset[0]["audio"]["array"]74        output = audio_classifier(audio)75        self.assertEqual(76            output,77            [78                {"score": ANY(float), "label": ANY(str)},79                {"score": ANY(float), "label": ANY(str)},80            ],81        )82 83    @require_torch84    def test_small_model_pt(self):85        model = "anton-l/wav2vec2-random-tiny-classifier"86 87        audio_classifier = pipeline("audio-classification", model=model)88 89        audio = np.ones((8000,))90        output = audio_classifier(audio, top_k=4)91 92        EXPECTED_OUTPUT = [93            {"score": 0.0842, "label": "no"},94            {"score": 0.0838, "label": "up"},95            {"score": 0.0837, "label": "go"},96            {"score": 0.0834, "label": "right"},97        ]98        EXPECTED_OUTPUT_PT_2 = [99            {"score": 0.0845, "label": "stop"},100            {"score": 0.0844, "label": "on"},101            {"score": 0.0841, "label": "right"},102            {"score": 0.0834, "label": "left"},103        ]104        self.assertIn(nested_simplify(output, decimals=4), [EXPECTED_OUTPUT, EXPECTED_OUTPUT_PT_2])105 106    @require_torch107    @slow108    def test_large_model_pt(self):109        import datasets110 111        model = "superb/wav2vec2-base-superb-ks"112 113        audio_classifier = pipeline("audio-classification", model=model)114        dataset = datasets.load_dataset("anton-l/superb_dummy", "ks", split="test")115 116        audio = np.array(dataset[3]["speech"], dtype=np.float32)117        output = audio_classifier(audio, top_k=4)118        self.assertEqual(119            nested_simplify(output, decimals=3),120            [121                {"score": 0.981, "label": "go"},122                {"score": 0.007, "label": "up"},123                {"score": 0.006, "label": "_unknown_"},124                {"score": 0.001, "label": "down"},125            ],126        )127 128    @require_tf129    @unittest.skip("Audio classification is not implemented for TF")130    def test_small_model_tf(self):131        pass132