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

[Speech Examples] Add pytorch speech pretraining (#13877)

Browse Source 
* adapt wav2vec2

* add example

* add files

* adapt

* remove bogus file

* Apply suggestions from code review

* adapt files more

* upload changes

* del old files

* up

* up

* up

* up

* up

* correct gradient checkpoitning

* add readme

* finish

* finish

* up

* more fixes

* up

* up

* add demo run to readme

* up
This commit is contained in:
Patrick von Platen
2021-10-12 00:46:32 +02:00
committed by GitHub
parent 3499728dc4
commit d45fc7da3d
9 changed files with 1196 additions and 183 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
tests/test_modeling_hubert.py
View File

@@ -586,7 +586,8 @@ class HubertUtilsTest(unittest.TestCase):
mask_prob = 0.5
mask_length = 1
mask = _compute_mask_indices((batch_size, sequence_length), mask_prob, mask_length, torch_device)
mask = _compute_mask_indices((batch_size, sequence_length), mask_prob, mask_length)
mask = torch.from_numpy(mask).to(torch_device)
self.assertListEqual(mask.sum(axis=-1).tolist(), [mask_prob * sequence_length for _ in range(batch_size)])
@@ -596,7 +597,8 @@ class HubertUtilsTest(unittest.TestCase):
mask_prob = 0.5
mask_length = 4
mask = _compute_mask_indices((batch_size, sequence_length), mask_prob, mask_length, torch_device)
mask = _compute_mask_indices((batch_size, sequence_length), mask_prob, mask_length)
mask = torch.from_numpy(mask).to(torch_device)
# because of overlap mask don't have to add up exactly to `mask_prob * sequence_length`, but have to be smaller or equal
for batch_sum in mask.sum(axis=-1):