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

[refactor] set attention implementation (#38974)

Browse Source 
* update

* fix some tests

* init from config, changes it in-place, add deepcopy in tests

* fix modernbert

* don't delete thsi config attr

* update

* style and copies

* skip tests in generation

* fix style

* accidentally removed flash-attn-3, revert

* docs

* forgot about flags set to False

* fix copies

* address a few comments

* fix copies

* custom code BC
This commit is contained in:
Raushan Turganbay
2025-07-15 12:34:06 +05:00
committed by GitHub
parent 6017f5e8ed
commit 8d6259b0b8
185 changed files with 451 additions and 776 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
tests/models/blip_2/test_modeling_blip_2.py
View File

@@ -13,6 +13,7 @@
# limitations under the License.
"""Testing suite for the PyTorch BLIP-2 model."""
import copy
import inspect
import tempfile
import unittest
@@ -184,9 +185,8 @@ class Blip2VisionModelTest(ModelTesterMixin, unittest.TestCase):
self.assertTrue(x is None or isinstance(x, nn.Linear))
def test_forward_signature(self):
config, _ = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
config, _ = self.model_tester.prepare_config_and_inputs_for_common()
model = model_class(config)
signature = inspect.signature(model.forward)
# signature.parameters is an OrderedDict => so arg_names order is deterministic
@@ -987,9 +987,8 @@ class Blip2ModelTest(ModelTesterMixin, PipelineTesterMixin, GenerationTesterMixi
raise ValueError("The eager model should not have SDPA attention layers")
def test_forward_signature(self):
config, _ = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
config, _ = self.model_tester.prepare_config_and_inputs_for_common()
model = model_class(config)
signature = inspect.signature(model.forward)
# signature.parameters is an OrderedDict => so arg_names order is deterministic
@@ -1077,7 +1076,7 @@ class Blip2ModelTest(ModelTesterMixin, PipelineTesterMixin, GenerationTesterMixi
for key in ["vision_config", "qformer_config", "text_config"]:
setattr(configs_no_init, key, _config_zero_init(getattr(configs_no_init, key)))
for model_class in self.all_model_classes:
model = model_class(config=configs_no_init)
model = model_class(config=copy.deepcopy(configs_no_init))
for name, param in model.named_parameters():
if param.requires_grad:
self.assertIn(