Add new meta w2v2-conformer BERT-like model #28165
ylacombe
merged
77 commits into
huggingface:main
from
ylacombe:add-w2v2-shaw
first commit
96bd9b78
correct default value non causal
d45e0751
update config and modeling code
35193abc
update converting checkpoint
730ee47d
clean modeling and fix tests
ad752605
make style
95c6542f
Merge branch 'huggingface:main' into add-w2v2-shaw
7f52d837
add new config parameters to docstring
175f4d3b
fix copied from statements
40c33570
sanchit-gandhi
commented
on 2023-12-20
Apply suggestions from code review
29e20ea1
make position_embeddings_type docstrings clearer
07b34ff8
clean converting script
c1f6ac4c
remove function not used
9792399c
clean modeling file
d267a7ca
apply suggestion for test file + add convert script to not_doctested
d69d850e
sanchit-gandhi
approved these changes
on 2023-12-21
modify tests according to review - cleaner logic and more tests
ca23f4ef
Apply nit suggestions from code review
c7550cd5
add checker of valid position embeddings type
1b132c65
instantiate new layer norm layer with the right eps
e179499b
fix freeze_feature_encoder since it can be None in some cases
4848a26c
add test same output in convert script
0a8e2a9e
restore wav2vec2conformer and add new model
5d77490d
create processor and FE + clean
b5b0bd2b
add new model code
aecc8fe8
fix convert script and set default config parameters
1d781003
correct model id paths
f98e5ab0
make style
b8a386aa
make fix-copies and cleaning files
580fa3e7
fix copied from statements
a847b08c
complete .md and fixe copies
0ededce7
clean convert script argument defaults
c0e92e2d
fix config parameters docstrings
43dd3794
fix config docstring
61162a76
Merge branch 'huggingface:main' into add-w2v2-shaw
ae706201
add copied from and enrich FE tests
ad0b1a97
fix copied from and repo-consistency
1d37cbbe
add autotokenizer
73bf2865
make test input length shorter and change docstring code
ca5ec51a
fix docstrings and copied from
dbc53ab8
add add_adapter to ASR training example
b18ad626
make testing of adapters more robust
7d856792
adapt to multi adapter layers
2ea024d0
refactor input_values->input_features and remove w2v2-bert feature ex…
bcdd067c
remove pretraining model
a17d2fa7
remove depreciated features and useless lines
2e6063db
add copied from and ignore statements to modeling tests
78ef614e
remove pretraining model #2
8a838314
change import in convert script
22c552f7
change default in convert script
91156155
update readme and remove useless line
2b570968
ylacombe
commented
on 2024-01-10
Update tests/models/wav2vec2_bert/test_processor_wav2vec2_bert.py
7db6c400
refactor BERT to Bert for consistency
87876a51
remove useless ignore copy statement
aaf64c16
add persistent to buffer in rotary
39b4ef13
add eps in LayerNorm init and remove copied from
7540037f
add adapter activation parameters and add copied from statements
14f5fad5
Fix copied statements and add unitest.skip reasons
85e88655
add copied statement in test_processor
2efa94bf
refactor processor
42ee5ae3
make style
f347b14f
replace numpy random by torch rand
632272ae
remove expected output CTC
0876dc94
Merge branch 'huggingface:main' into add-w2v2-shaw
8aa7bdfa
improve converting script with processor class
2f4cff38
Apply suggestions from code review
18567101
remove gumbel class
a0e5ca2e
remove tests related to previously deleted class
b5e7d70b
Update src/transformers/models/wav2vec2_bert/configuration_wav2vec2_b…
fc6dfc2e
correct typos
ef1ad917
remove uused parameters
f3d00e62
update processor to takes both text and audio
849f3f50
update checkpoints
72667100
update expected output and add ctc expected output
07de02ec
add label_attention_mask
37dd941d
replace pt with np in processor tests
8b37745b
amyeroberts
commented
on 2024-01-17
fix typo
71609062
revert to behaviour with labels_attention_mask
c6ec6b17
amyeroberts
approved these changes
on 2024-01-18
Reviewers
Assignees
No one assigned
Labels
None yet
Milestone
No milestone
Login to write a write a comment.
Login via GitHub