Add TF port of BLIP #22090
Rocketknight1
merged
57 commits into
main
from
add_tf_blip
gante
commented
on 2023-03-29
Initial commit
98b2afda
more stash commit
e557c346
Yet another stash commit
87767b0a
yet more stash commit
d86ec346
Mostly working except for docs / repo consistency
35deb284
Stop importing model list from torch file
0a720e4c
Add TF BLIP models to docs
490fc637
Add auto classes
6dc06bbc
Move get_text_features and get_image_features
9fd4b76b
Update src/transformers/models/blip/modeling_tf_blip.py
07f99eb5
Update src/transformers/models/blip/modeling_tf_blip.py
8cfc37df
Update src/transformers/models/blip/modeling_tf_blip.py
1c47a2f1
Update src/transformers/models/blip/modeling_tf_blip_text.py
70cfe55e
Update src/transformers/models/blip/modeling_tf_blip.py
2024f5ec
Update src/transformers/models/blip/modeling_tf_blip.py
cc1694db
Update src/transformers/models/blip/modeling_tf_blip.py
f31e96bd
Update src/transformers/models/blip/modeling_tf_blip.py
e12e3052
Update src/transformers/models/blip/modeling_tf_blip.py
2d622f66
Update tests/models/blip/test_modeling_tf_blip.py
ad2c87cc
Update tests/models/blip/test_modeling_tf_blip.py
6b781dfb
Update src/transformers/models/blip/modeling_tf_blip.py
dab565b7
Update src/transformers/models/blip/modeling_tf_blip.py
ee823fcd
Update tests/models/blip/test_modeling_tf_blip_text.py
d6c58694
Update src/transformers/models/blip/modeling_tf_blip_text.py
cf307fa4
Update src/transformers/models/blip/modeling_tf_blip.py
0289c286
Use channels_last convolutions in TF (better performance + compatibil…
c4a4b62d
Remove _shape function
3a082f8c
Move multi-line statement to one line in PT + TF
8e73e08f
Specify tf.keras.layers instead of importing from it
7d0f73b2
Remove test_gradient_checkpointing and empty test_training methods
4ec371b7
move some multi-line statements to one line
561d2f8b
Update docstring for generate
076948be
Remove pruned heads set
429c25ee
Remove self.seq_len_dim
30862572
Fixed issues with loss computation, should resolve some tests. Also e…
adb03307
ensure original model follows config in more cases
fba23856
Skip the same cross-attention tests in the PT tests - didn't realize …
f6c328ed
Add training args throughout the models and layers
4d71a053
make fixup
7239db57
Fix docstring for inputs_embeds
09592b22
Add docstring for is_decoder
d4a6fa60
Add docstrings to text models
60f078c6
Remove redundant computation
e6a7851c
Add unpack_inputs / keras_serializable
f3062b14
Add modeling_tf_blip to doctests
77e365ee
Add config classes for keras serialization
6fff45c9
gante
approved these changes
on 2023-03-30
Changes to allow model porting with pt-to-tf
34463eae
Quick fix to decoder head and test tweaks
60b7fb7e
Revert an issue with masking the embeddings outputs
2a7f52de
Allow missing keys in some equivalence tests (for unused layers)
d962ac6b
Add tf-pt equivalence tests back in
0a43f85c
Update src/transformers/models/blip/modeling_tf_blip.py
09095d1c
Update src/transformers/models/blip/modeling_tf_blip_text.py
dd88c83b
Update src/transformers/models/blip/modeling_tf_blip_text.py
d0fd3d4b
make fixup
9efd53c1
Refactor invert_attention_mask out into tf_utils
afd5a9cf
sgugger
commented
on 2023-04-03
Re-enable cross-tests on the PT side too
41fe5e19
sgugger
approved these changes
on 2023-04-04
gante
approved these changes
on 2023-04-04
Reviewers
Assignees
No one assigned
Labels
None yet
Milestone
No milestone
Login to write a write a comment.
Login via GitHub