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

Move TF building to an actual build() method (#23760)

Browse Source 
* A fun new PR where I break the entire codebase again

* A fun new PR where I break the entire codebase again

* Handle cross-attention

* Move calls to model(model.dummy_inputs) to the new build() method

* Seeing what fails with the build context thing

* make fix-copies

* Let's see what fails with new build methods

* Fix the pytorch crossload build calls

* Fix the overridden build methods in vision_text_dual_encoder

* Make sure all our build methods set self.built or call super().build(), which also sets it

* make fix-copies

* Remove finished TODO

* Tentatively remove unneeded (?) line

* Transpose b in deberta correctly and remove unused threading local

* Get rid of build_with_dummies and all it stands for

* Rollback some changes to TF-PT crossloading

* Correctly call super().build()
This commit is contained in:
Matt
2023-06-06 18:30:51 +01:00
committed by GitHub
parent cbf6bc2350
commit 4a55e47877
27 changed files with 159 additions and 138 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
tests/test_modeling_tf_common.py
View File

@@ -348,7 +348,7 @@ class TFModelTesterMixin:
with tf.Graph().as_default() as g:
model = model_class(config)
model(model.dummy_inputs)
model.build()
for op in g.get_operations():
model_op_names.add(op.node_def.op)
@@ -375,7 +375,7 @@ class TFModelTesterMixin:
for model_class in self.all_model_classes:
model = model_class(config)
model(model.dummy_inputs)
model.build()
onnx_model_proto, _ = tf2onnx.convert.from_keras(model, opset=self.onnx_min_opset)
@@ -1180,7 +1180,7 @@ class TFModelTesterMixin:
def _get_word_embedding_weight(model, embedding_layer):
if isinstance(embedding_layer, tf.keras.layers.Embedding):
# builds the embeddings layer
model(model.dummy_inputs)
model.build()
return embedding_layer.embeddings
else:
return model._get_word_embedding_weight(embedding_layer)
@@ -1243,7 +1243,7 @@ class TFModelTesterMixin:
old_total_size = config.vocab_size
new_total_size = old_total_size + new_tokens_size
model = model_class(config=copy.deepcopy(config)) # `resize_token_embeddings` mutates `config`
model(model.dummy_inputs) # builds the embeddings layer
model.build()
model.resize_token_embeddings(new_total_size)
# fetch the output for an input exclusively made of new members of the vocabulary
@@ -2313,8 +2313,8 @@ class UtilsFunctionsTest(unittest.TestCase):
# Finally, check the model can be reloaded
new_model = TFBertModel.from_pretrained(tmp_dir)
model(model.dummy_inputs)
new_model(model.dummy_inputs)
model.build()
new_model.build()
for p1, p2 in zip(model.weights, new_model.weights):
self.assertTrue(np.allclose(p1.numpy(), p2.numpy()))
@@ -2440,7 +2440,7 @@ class TFModelPushToHubTester(unittest.TestCase):
)
model = TFBertModel(config)
# Make sure model is properly initialized
_ = model(model.dummy_inputs)
model.build()
logging.set_verbosity_info()
logger = logging.get_logger("transformers.utils.hub")
@@ -2509,7 +2509,7 @@ class TFModelPushToHubTester(unittest.TestCase):
)
model = TFBertModel(config)
# Make sure model is properly initialized
_ = model(model.dummy_inputs)
model.build()
model.push_to_hub("valid_org/test-model-tf-org", use_auth_token=self._token)