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

Allow user-managed Pool in Wav2Vec2ProcessorWithLM.batch_decode (#18351)

Browse Source 
* [Wav2Vec2] Allow user-managed Pool in Wav2Vec2ProcessorWithLM.batch_decode

* [Wav2Vec2] Add user-managed LM's pool tests and usage examples

* Improve styling

Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* [Wav2Vec2] Fix hyperlink references

Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
This commit is contained in:
Antonio Carlos Falcão Petri
2022-10-18 09:48:03 -03:00
committed by GitHub
parent bf0e094142
commit af150e4a1c
10 changed files with 348 additions and 63 deletions
Download Patch File Download Diff File Expand all files Collapse all files

57
tests/models/wav2vec2/test_modeling_tf_wav2vec2.py
View File

@@ -18,6 +18,7 @@ import copy
import glob
import inspect
import math
import multiprocessing
import unittest
import numpy as np
@@ -26,7 +27,7 @@ from datasets import load_dataset
from huggingface_hub import snapshot_download
from transformers import Wav2Vec2Config, is_tf_available
from transformers.testing_utils import is_flaky, require_librosa, require_pyctcdecode, require_tf, slow
from transformers.testing_utils import CaptureLogger, is_flaky, require_librosa, require_pyctcdecode, require_tf, slow
from transformers.utils import is_librosa_available, is_pyctcdecode_available
from ...test_configuration_common import ConfigTester
@@ -42,6 +43,7 @@ if is_tf_available():
if is_pyctcdecode_available():
from transformers import Wav2Vec2ProcessorWithLM
from transformers.models.wav2vec2_with_lm import processing_wav2vec2_with_lm
if is_librosa_available():
@@ -590,3 +592,56 @@ class TFWav2Vec2ModelIntegrationTest(unittest.TestCase):
transcription = processor.batch_decode(logits.numpy()).text
self.assertEqual(transcription[0], "el libro ha sido escrito por cervantes")
@require_pyctcdecode
@require_librosa
def test_wav2vec2_with_lm_pool(self):
downloaded_folder = snapshot_download("patrickvonplaten/common_voice_es_sample")
file_path = glob.glob(downloaded_folder + "/*")[0]
sample = librosa.load(file_path, sr=16_000)[0]
model = TFWav2Vec2ForCTC.from_pretrained("patrickvonplaten/wav2vec2-large-xlsr-53-spanish-with-lm")
processor = Wav2Vec2ProcessorWithLM.from_pretrained("patrickvonplaten/wav2vec2-large-xlsr-53-spanish-with-lm")
input_values = processor(sample, return_tensors="tf").input_values
logits = model(input_values).logits
# test user-managed pool
with multiprocessing.get_context("fork").Pool(2) as pool:
transcription = processor.batch_decode(logits.numpy(), pool).text
self.assertEqual(transcription[0], "el libro ha sido escrito por cervantes")
# user-managed pool + num_processes should trigger a warning
with CaptureLogger(processing_wav2vec2_with_lm.logger) as cl, multiprocessing.get_context("fork").Pool(
2
) as pool:
transcription = processor.batch_decode(logits.numpy(), pool, num_processes=2).text
self.assertIn("num_process", cl.out)
self.assertIn("it will be ignored", cl.out)
self.assertEqual(transcription[0], "el libro ha sido escrito por cervantes")
@require_pyctcdecode
@require_librosa
def test_wav2vec2_with_lm_invalid_pool(self):
downloaded_folder = snapshot_download("patrickvonplaten/common_voice_es_sample")
file_path = glob.glob(downloaded_folder + "/*")[0]
sample = librosa.load(file_path, sr=16_000)[0]
model = TFWav2Vec2ForCTC.from_pretrained("patrickvonplaten/wav2vec2-large-xlsr-53-spanish-with-lm")
processor = Wav2Vec2ProcessorWithLM.from_pretrained("patrickvonplaten/wav2vec2-large-xlsr-53-spanish-with-lm")
input_values = processor(sample, return_tensors="tf").input_values
logits = model(input_values).logits
# change default start method, which should trigger a warning if different than fork
multiprocessing.set_start_method("spawn")
with CaptureLogger(processing_wav2vec2_with_lm.logger) as cl:
transcription = processor.batch_decode(logits.numpy()).text
self.assertIn("Falling back to sequential decoding.", cl.out)
self.assertEqual(transcription[0], "el libro ha sido escrito por cervantes")