riaz.somc/transformers
1
0
Fork 0
mirror of https://github.com/huggingface/transformers.git synced 2025-07-03 04:40:06 +06:00
Code Issues Actions 28 Packages Projects Releases Wiki Activity

🚨Early-error🚨 config will error out if output_attentions=True and the attn implementation is wrong (#38288)

Browse Source 
* Protect ParallelInterface

* early error out on output attention setting for no wraning in modeling

* modular update

* fixup

* update model tests

* update

* oups

* set model's config

* more cases

* ??

* properly fix

* fixup

* update

* last onces

* update

* fix?

* fix wrong merge commit

* fix hub test

* nits

* wow I am tired

* updates

* fix pipeline!

---------

Co-authored-by: Lysandre <hi@lysand.re>
This commit is contained in:
Arthur 2025-05-23 17:17:38 +02:00 committed by GitHub
parent 896833c183
commit f5d45d89c4
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
71 changed files with 157 additions and 144 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
docs/source/en/model_doc/jamba.md
View File

@ -99,7 +99,7 @@ quantization_config = BitsAndBytesConfig(load_in_8bit=True,
device_map = {'model.embed_tokens': 0, 'model.layers.0': 0, 'model.layers.1': 0, 'model.layers.2': 0, 'model.layers.3': 0, 'model.layers.4': 0, 'model.layers.5': 0, 'model.layers.6': 0, 'model.layers.7': 0, 'model.layers.8': 0, 'model.layers.9': 1, 'model.layers.10': 1, 'model.layers.11': 1, 'model.layers.12': 1, 'model.layers.13': 1, 'model.layers.14': 1, 'model.layers.15': 1, 'model.layers.16': 1, 'model.layers.17': 1, 'model.layers.18': 2, 'model.layers.19': 2, 'model.layers.20': 2, 'model.layers.21': 2, 'model.layers.22': 2, 'model.layers.23': 2, 'model.layers.24': 2, 'model.layers.25': 2, 'model.layers.26': 2, 'model.layers.27': 3, 'model.layers.28': 3, 'model.layers.29': 3, 'model.layers.30': 3, 'model.layers.31': 3, 'model.layers.32': 3, 'model.layers.33': 3, 'model.layers.34': 3, 'model.layers.35': 3, 'model.layers.36': 4, 'model.layers.37': 4, 'model.layers.38': 4, 'model.layers.39': 4, 'model.layers.40': 4, 'model.layers.41': 4, 'model.layers.42': 4, 'model.layers.43': 4, 'model.layers.44': 4, 'model.layers.45': 5, 'model.layers.46': 5, 'model.layers.47': 5, 'model.layers.48': 5, 'model.layers.49': 5, 'model.layers.50': 5, 'model.layers.51': 5, 'model.layers.52': 5, 'model.layers.53': 5, 'model.layers.54': 6, 'model.layers.55': 6, 'model.layers.56': 6, 'model.layers.57': 6, 'model.layers.58': 6, 'model.layers.59': 6, 'model.layers.60': 6, 'model.layers.61': 6, 'model.layers.62': 6, 'model.layers.63': 7, 'model.layers.64': 7, 'model.layers.65': 7, 'model.layers.66': 7, 'model.layers.67': 7, 'model.layers.68': 7, 'model.layers.69': 7, 'model.layers.70': 7, 'model.layers.71': 7, 'model.final_layernorm': 7, 'lm_head': 7}
model = AutoModelForCausalLM.from_pretrained("ai21labs/AI21-Jamba-Large-1.6",
torch_dtype=torch.bfloat16,
attn_implementation="flash_attention_2",
attn_implementation="flash_attention_2",
quantization_config=quantization_config,
device_map=device_map)

20
src/transformers/configuration_utils.py
View File

@ -214,7 +214,7 @@ class PretrainedConfig(PushToHubMixin):
# Attributes with defaults
self.return_dict = kwargs.pop("return_dict", True)
self.output_hidden_states = kwargs.pop("output_hidden_states", False)
self.output_attentions = kwargs.pop("output_attentions", False)
self._output_attentions = kwargs.pop("output_attentions", False)
self.torchscript = kwargs.pop("torchscript", False) # Only used by PyTorch models
self.torch_dtype = kwargs.pop("torch_dtype", None) # Only used by PyTorch models
self.use_bfloat16 = kwargs.pop("use_bfloat16", False)
@ -331,6 +331,22 @@ class PretrainedConfig(PushToHubMixin):
def name_or_path(self, value):
self._name_or_path = str(value) # Make sure that name_or_path is a string (for JSON encoding)
@property
def output_attentions(self):
"""
`bool`: Whether or not the model should returns all attentions.
"""
return self._output_attentions
@output_attentions.setter
def output_attentions(self, value):
if self._attn_implementation != "eager":
raise ValueError(
"The `output_attentions` attribute is not supported when using the `attn_implementation` set to "
f"{self._attn_implementation}. Please set it to 'eager' instead."
)
self._output_attentions = value
@property
def use_return_dict(self) -> bool:
"""
@ -1004,6 +1020,8 @@ class PretrainedConfig(PushToHubMixin):
if "_auto_class" in d:
del d["_auto_class"]
if "_output_attentions" in d:
d["output_attentions"] = d.pop("_output_attentions")
if "_commit_hash" in d:
del d["_commit_hash"]
if "_attn_implementation_internal" in d:

9
src/transformers/models/aria/modeling_aria.py
View File

@ -549,15 +549,8 @@ class AriaTextAttention(nn.Module):
key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs)
attention_interface: Callable = eager_attention_forward
if self.config._attn_implementation != "eager":
if self.config._attn_implementation == "sdpa" and kwargs.get("output_attentions", False):
logger.warning_once(
"`torch.nn.functional.scaled_dot_product_attention` does not support `output_attentions=True`. Falling back to "
'eager attention. This warning can be removed using the argument `attn_implementation="eager"` when loading the model.'
)
else:
attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
attn_output, attn_weights = attention_interface(
self,

9
src/transformers/models/bamba/modeling_bamba.py
View File

@ -313,15 +313,8 @@ class BambaAttention(nn.Module):
key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs)
attention_interface: Callable = eager_attention_forward
if self.config._attn_implementation != "eager":
if self.config._attn_implementation == "sdpa" and kwargs.get("output_attentions", False):
logger.warning_once(
"`torch.nn.functional.scaled_dot_product_attention` does not support `output_attentions=True`. Falling back to "
'eager attention. This warning can be removed using the argument `attn_implementation="eager"` when loading the model.'
)
else:
attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
attn_output, attn_weights = attention_interface(
self,

9
src/transformers/models/csm/modeling_csm.py
View File

@ -337,15 +337,8 @@ class CsmAttention(nn.Module):
key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs)
attention_interface: Callable = eager_attention_forward
if self.config._attn_implementation != "eager":
if self.config._attn_implementation == "sdpa" and kwargs.get("output_attentions", False):
logger.warning_once(
"`torch.nn.functional.scaled_dot_product_attention` does not support `output_attentions=True`. Falling back to "
'eager attention. This warning can be removed using the argument `attn_implementation="eager"` when loading the model.'
)
else:
attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
attn_output, attn_weights = attention_interface(
self,

9
src/transformers/models/emu3/modeling_emu3.py
View File

@ -206,15 +206,8 @@ class Emu3Attention(nn.Module):
key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs)
attention_interface: Callable = eager_attention_forward
if self.config._attn_implementation != "eager":
if self.config._attn_implementation == "sdpa" and kwargs.get("output_attentions", False):
logger.warning_once(
"`torch.nn.functional.scaled_dot_product_attention` does not support `output_attentions=True`. Falling back to "
'eager attention. This warning can be removed using the argument `attn_implementation="eager"` when loading the model.'
)
else:
attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
attn_output, attn_weights = attention_interface(
self,

9
src/transformers/models/gemma/modeling_gemma.py
View File

@ -239,15 +239,8 @@ class GemmaAttention(nn.Module):
key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs)
attention_interface: Callable = eager_attention_forward
if self.config._attn_implementation != "eager":
if self.config._attn_implementation == "sdpa" and kwargs.get("output_attentions", False):
logger.warning_once(
"`torch.nn.functional.scaled_dot_product_attention` does not support `output_attentions=True`. Falling back to "
'eager attention. This warning can be removed using the argument `attn_implementation="eager"` when loading the model.'
)
else:
attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
attn_output, attn_weights = attention_interface(
self,

9
src/transformers/models/glm/modeling_glm.py
View File

@ -201,15 +201,8 @@ class GlmAttention(nn.Module):
key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs)
attention_interface: Callable = eager_attention_forward
if self.config._attn_implementation != "eager":
if self.config._attn_implementation == "sdpa" and kwargs.get("output_attentions", False):
logger.warning_once(
"`torch.nn.functional.scaled_dot_product_attention` does not support `output_attentions=True`. Falling back to "
'eager attention. This warning can be removed using the argument `attn_implementation="eager"` when loading the model.'
)
else:
attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
attn_output, attn_weights = attention_interface(
self,

9
src/transformers/models/glm4/modeling_glm4.py
View File

@ -259,15 +259,8 @@ class Glm4Attention(nn.Module):
key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs)
attention_interface: Callable = eager_attention_forward
if self.config._attn_implementation != "eager":
if self.config._attn_implementation == "sdpa" and kwargs.get("output_attentions", False):
logger.warning_once(
"`torch.nn.functional.scaled_dot_product_attention` does not support `output_attentions=True`. Falling back to "
'eager attention. This warning can be removed using the argument `attn_implementation="eager"` when loading the model.'
)
else:
attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
attn_output, attn_weights = attention_interface(
self,

9
src/transformers/models/granite/modeling_granite.py
View File

@ -165,15 +165,8 @@ class GraniteAttention(nn.Module):
key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs)
attention_interface: Callable = eager_attention_forward
if self.config._attn_implementation != "eager":
if self.config._attn_implementation == "sdpa" and kwargs.get("output_attentions", False):
logger.warning_once(
"`torch.nn.functional.scaled_dot_product_attention` does not support `output_attentions=True`. Falling back to "
'eager attention. This warning can be removed using the argument `attn_implementation="eager"` when loading the model.'
)
else:
attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
attn_output, attn_weights = attention_interface(
self,

9
src/transformers/models/helium/modeling_helium.py
View File

@ -241,15 +241,8 @@ class HeliumAttention(nn.Module):
key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs)
attention_interface: Callable = eager_attention_forward
if self.config._attn_implementation != "eager":
if self.config._attn_implementation == "sdpa" and kwargs.get("output_attentions", False):
logger.warning_once(
"`torch.nn.functional.scaled_dot_product_attention` does not support `output_attentions=True`. Falling back to "
'eager attention. This warning can be removed using the argument `attn_implementation="eager"` when loading the model.'
)
else:
attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
attn_output, attn_weights = attention_interface(
self,

9
src/transformers/models/llama/modeling_llama.py
View File

@ -245,15 +245,8 @@ class LlamaAttention(nn.Module):
key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs)
attention_interface: Callable = eager_attention_forward
if self.config._attn_implementation != "eager":
if self.config._attn_implementation == "sdpa" and kwargs.get("output_attentions", False):
logger.warning_once(
"`torch.nn.functional.scaled_dot_product_attention` does not support `output_attentions=True`. Falling back to "
'eager attention. This warning can be removed using the argument `attn_implementation="eager"` when loading the model.'
)
else:
attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
attn_output, attn_weights = attention_interface(
self,

7
tests/generation/test_utils.py
View File

@ -1157,7 +1157,8 @@ class GenerationTesterMixin:
self.skipTest(reason=f"{model_class.__name__} doesn't support caching")
config.is_decoder = True
model = model_class(config).to(torch_device).eval()
model = model_class._from_config(config, attn_implementation="eager").to(torch_device).eval()
config = model.config
# Sets assisted generation arguments such that:
# a) no EOS is generated, to ensure generation doesn't break early
# b) the assistant model always generates two tokens when it is called, to ensure the input preparation of
@ -1187,6 +1188,7 @@ class GenerationTesterMixin:
assistant_model = model_class(config).to(torch_device).eval()
else:
assistant_model = model
assistant_model.config._attn_implementation = "eager"
assistant_model.generation_config.num_assistant_tokens = 2 # see b)
assistant_model.generation_config.num_assistant_tokens_schedule = "constant" # see b)
generation_kwargs.update({"assistant_model": assistant_model})
@ -1367,7 +1369,8 @@ class GenerationTesterMixin:
self.skipTest(reason=f"{model_class.__name__} doesn't support caching")
config.is_decoder = True
model = model_class(config).to(torch_device).eval()
model = model_class._from_config(config, attn_implementation="eager").to(torch_device).eval()
config = model.config
# Sets assisted generation arguments such that:
# a) no EOS is generated, to ensure generation doesn't break early
# b) the assistant model always generates two tokens when it is called, to ensure the input preparation of

3
tests/models/autoformer/test_modeling_autoformer.py
View File

@ -323,7 +323,8 @@ class AutoformerModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCa
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/bamba/test_modeling_bamba.py
View File

@ -362,7 +362,8 @@ class BambaModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixi
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()

3
tests/models/canine/test_modeling_canine.py
View File

@ -318,7 +318,8 @@ class CanineModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/conditional_detr/test_modeling_conditional_detr.py
View File

@ -279,7 +279,8 @@ class ConditionalDetrModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.T
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/convbert/test_modeling_convbert.py
View File

@ -327,7 +327,8 @@ class ConvBertModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/d_fine/test_modeling_d_fine.py
View File

@ -371,7 +371,8 @@ class DFineModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/dab_detr/test_modeling_dab_detr.py
View File

@ -499,7 +499,8 @@ class DabDetrModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase)
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/deformable_detr/test_modeling_deformable_detr.py
View File

@ -278,7 +278,8 @@ class DeformableDetrModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.Te
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/detr/test_modeling_detr.py
View File

@ -279,7 +279,8 @@ class DetrModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/donut/test_modeling_donut_swin.py
View File

@ -214,7 +214,8 @@ class DonutSwinModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCas
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

1
tests/models/encoder_decoder/test_modeling_encoder_decoder.py
View File

@ -457,6 +457,7 @@ class EncoderDecoderMixin:
decoder_attention_mask = decoder_attention_mask[:, :-1]
encoder_model, decoder_model = self.get_encoder_decoder_model(config, decoder_config)
enc_dec_model = EncoderDecoderModel(encoder=encoder_model, decoder=decoder_model)
enc_dec_model.config._attn_implementation = "eager" # model config -> won't work
enc_dec_model.config.output_attentions = True # model config -> won't work
enc_dec_model.to(torch_device)
outputs_encoder_decoder = enc_dec_model(

3
tests/models/falcon_h1/test_modeling_falcon_h1.py
View File

@ -337,7 +337,8 @@ class FalconH1ModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterM
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()

6
tests/models/fastspeech2_conformer/test_modeling_fastspeech2_conformer.py
View File

@ -287,7 +287,8 @@ class FastSpeech2ConformerModelTest(ModelTesterMixin, unittest.TestCase):
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():
@ -709,7 +710,8 @@ class FastSpeech2ConformerWithHifiGanTest(ModelTesterMixin, unittest.TestCase):
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.model_config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/flava/test_modeling_flava.py
View File

@ -218,7 +218,8 @@ class FlavaImageModelTest(ModelTesterMixin, unittest.TestCase):
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/glpn/test_modeling_glpn.py
View File

@ -184,7 +184,8 @@ class GLPNModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/grounding_dino/test_modeling_grounding_dino.py
View File

@ -327,7 +327,8 @@ class GroundingDinoModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.Tes
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/groupvit/test_modeling_groupvit.py
View File

@ -202,7 +202,8 @@ class GroupViTVisionModelTest(ModelTesterMixin, unittest.TestCase):
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/hiera/test_modeling_hiera.py
View File

@ -281,7 +281,8 @@ class HieraModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/idefics/test_modeling_idefics.py
View File

@ -502,7 +502,8 @@ class IdeficsModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase)
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/informer/test_modeling_informer.py
View File

@ -384,7 +384,8 @@ class InformerModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/jamba/test_modeling_jamba.py
View File

@ -452,7 +452,8 @@ class JambaModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixi
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()

3
tests/models/layoutlmv2/test_modeling_layoutlmv2.py
View File

@ -334,7 +334,8 @@ class LayoutLMv2ModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCa
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/led/test_modeling_led.py
View File

@ -405,7 +405,8 @@ class LEDModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixin,
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/luke/test_modeling_luke.py
View File

@ -758,7 +758,8 @@ class LukeModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/maskformer/test_modeling_maskformer.py
View File

@ -299,7 +299,8 @@ class MaskFormerModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCa
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/moonshine/test_modeling_moonshine.py
View File

@ -167,7 +167,8 @@ class MoonshineModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCas
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()

3
tests/models/omdet_turbo/test_modeling_omdet_turbo.py
View File

@ -462,7 +462,8 @@ class OmDetTurboModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCa
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/paligemma/test_modeling_paligemma.py
View File

@ -355,7 +355,8 @@ class PaliGemmaForConditionalGenerationModelTest(ModelTesterMixin, GenerationTes
config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()

3
tests/models/pegasus_x/test_modeling_pegasus_x.py
View File

@ -304,7 +304,8 @@ class PegasusXModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterM
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/perceiver/test_modeling_perceiver.py
View File

@ -456,7 +456,8 @@ class PerceiverModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCas
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/rt_detr/test_modeling_rt_detr.py
View File

@ -335,7 +335,8 @@ class RTDetrModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/rt_detr_v2/test_modeling_rt_detr_v2.py
View File

@ -339,7 +339,8 @@ class RTDetrV2ModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/rwkv/test_modeling_rwkv.py
View File

@ -312,7 +312,8 @@ class RwkvModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixin
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()

6
tests/models/sam/test_modeling_sam.py
View File

@ -202,7 +202,8 @@ class SamVisionModelTest(ModelTesterMixin, unittest.TestCase):
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():
@ -590,7 +591,8 @@ class SamModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/sam/test_modeling_tf_sam.py
View File

@ -562,7 +562,8 @@ class TFSamModelTest(TFModelTesterMixin, PipelineTesterMixin, unittest.TestCase)
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
outputs = model(**self._prepare_for_class(inputs_dict, model_class))
vision_attentions = outputs.vision_attentions

6
tests/models/sam_hq/test_modeling_sam_hq.py
View File

@ -210,7 +210,8 @@ class SamHQVisionModelTest(ModelTesterMixin, unittest.TestCase):
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():
@ -637,7 +638,8 @@ class SamHQModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/seamless_m4t/test_modeling_seamless_m4t.py
View File

@ -475,7 +475,8 @@ class SeamlessM4TModelWithSpeechInputTest(ModelTesterMixin, unittest.TestCase):
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/seamless_m4t_v2/test_modeling_seamless_m4t_v2.py
View File

@ -491,7 +491,8 @@ class SeamlessM4Tv2ModelWithSpeechInputTest(ModelTesterMixin, unittest.TestCase)
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/segformer/test_modeling_segformer.py
View File

@ -216,7 +216,8 @@ class SegformerModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCas
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/speech_to_text/test_modeling_speech_to_text.py
View File

@ -434,7 +434,8 @@ class Speech2TextModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTest
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()

3
tests/models/speech_to_text/test_modeling_tf_speech_to_text.py
View File

@ -323,7 +323,8 @@ class TFSpeech2TextModelTest(TFModelTesterMixin, PipelineTesterMixin, unittest.T
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
subsampled_encoder_seq_length = model._get_feat_extract_output_lengths(encoder_seq_length)
subsampled_encoder_key_length = model._get_feat_extract_output_lengths(encoder_key_length)

6
tests/models/speecht5/test_modeling_speecht5.py
View File

@ -415,7 +415,8 @@ class SpeechT5ForSpeechToTextTest(ModelTesterMixin, unittest.TestCase, Generatio
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
@ -1524,7 +1525,8 @@ class SpeechT5ForSpeechToSpeechTest(ModelTesterMixin, unittest.TestCase):
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()

3
tests/models/swin/test_modeling_swin.py
View File

@ -303,7 +303,8 @@ class SwinModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/swin2sr/test_modeling_swin2sr.py
View File

@ -263,7 +263,8 @@ class Swin2SRModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase)
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/swinv2/test_modeling_swinv2.py
View File

@ -286,7 +286,8 @@ class Swinv2ModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/table_transformer/test_modeling_table_transformer.py
View File

@ -293,7 +293,8 @@ class TableTransformerModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/time_series_transformer/test_modeling_time_series_transformer.py
View File

@ -286,7 +286,8 @@ class TimeSeriesTransformerModelTest(ModelTesterMixin, PipelineTesterMixin, unit
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/vilt/test_modeling_vilt.py
View File

@ -373,7 +373,8 @@ class ViltModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/visual_bert/test_modeling_visual_bert.py
View File

@ -412,7 +412,8 @@ class VisualBertModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCa
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/vivit/test_modeling_vivit.py
View File

@ -243,7 +243,8 @@ class VivitModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/whisper/test_modeling_whisper.py
View File

@ -676,7 +676,8 @@ class WhisperModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMi
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()

3
tests/models/x_clip/test_modeling_x_clip.py
View File

@ -236,7 +236,8 @@ class XCLIPVisionModelTest(ModelTesterMixin, unittest.TestCase):
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/yolos/test_modeling_yolos.py
View File

@ -233,7 +233,8 @@ class YolosModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():

3
tests/models/zamba/test_modeling_zamba.py
View File

@ -401,7 +401,8 @@ class ZambaModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixi
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()

3
tests/models/zamba2/test_modeling_zamba2.py
View File

@ -422,7 +422,8 @@ class Zamba2ModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMix
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()

1
tests/pipelines/test_pipelines_text_to_audio.py
View File

@ -259,6 +259,7 @@ class TextToAudioPipelineTests(unittest.TestCase):
model_test_kwargs = {}
if model.can_generate(): # not all models in this pipeline can generate and, therefore, take `generate` kwargs
model_test_kwargs["max_new_tokens"] = 5
model.config._attn_implementation = "eager"
speech_generator = TextToAudioPipeline(
model=model,
tokenizer=tokenizer,

5
tests/test_modeling_common.py
View File

@ -974,7 +974,8 @@ class ModelTesterMixin:
inputs_dict["output_attentions"] = True
inputs_dict["output_hidden_states"] = False
config.return_dict = True
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
config = model.config
model.to(torch_device)
model.eval()
with torch.no_grad():
@ -1720,7 +1721,7 @@ class ModelTesterMixin:
# no need to test all models as different heads yield the same functionality
model_class = self.all_model_classes[0]
model = model_class(config)
model = model_class._from_config(config, attn_implementation="eager")
model.to(torch_device)
inputs = self._prepare_for_class(inputs_dict, model_class)

1
tests/utils/test_configuration_utils.py
View File

@ -189,6 +189,7 @@ class ConfigTestUtils(unittest.TestCase):
self.assertListEqual(
missing_keys,
[
"_output_attentions",
"is_encoder_decoder",
"_name_or_path",
"_commit_hash",