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

make wav2vec2 test deterministic (#10714)

Browse Source
This commit is contained in:
Patrick von Platen
2021-03-15 16:50:05 +03:00
committed by GitHub
parent 6bef764506
commit d9e693e1d0
1 changed files with 4 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
tests/test_modeling_wav2vec2.py
View File

@@ -515,6 +515,8 @@ class Wav2Vec2ModelIntegrationTest(unittest.TestCase):
import soundfile as sf import soundfile as sf
ids = [f"1272-141231-000{i}" for i in range(num_samples)]
# map files to raw # map files to raw
def map_to_array(batch): def map_to_array(batch):
speech, _ = sf.read(batch["file"]) speech, _ = sf.read(batch["file"])
@@ -522,7 +524,8 @@ class Wav2Vec2ModelIntegrationTest(unittest.TestCase):
return batch return batch
ds = load_dataset("patrickvonplaten/librispeech_asr_dummy", "clean", split="validation") ds = load_dataset("patrickvonplaten/librispeech_asr_dummy", "clean", split="validation")
ds = ds.select(range(num_samples)).map(map_to_array)
ds = ds.filter(lambda x: x["id"] in ids).sort("id").map(map_to_array)
return ds["speech"][:num_samples] return ds["speech"][:num_samples]