dnrops
/
transformers
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Fix FA2 tests (#29909) 

* fix FA2 tests

* refactor inference test name
This commit is contained in:
Yoach Lacombe 2024-04-01 08:51:00 +01:00 committed by GitHub
parent 3b8e2932ce
commit 569f6c7d43
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
9 changed files with 15 additions and 19 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
tests/models/bark/test_modeling_bark.py
View File

@ -879,7 +879,7 @@ class BarkFineModelTest(ModelTesterMixin, unittest.TestCase):
@require_torch_gpu
@pytest.mark.flash_attn_test
@slow
def test_flash_attn_2_inference(self):
def test_flash_attn_2_inference_equivalence(self):
for model_class in self.all_model_classes:
if not model_class._supports_flash_attn_2:
return
@ -936,7 +936,7 @@ class BarkFineModelTest(ModelTesterMixin, unittest.TestCase):
@require_torch_gpu
@pytest.mark.flash_attn_test
@slow
def test_flash_attn_2_inference_padding_right(self):
def test_flash_attn_2_inference_equivalence_right_padding(self):
for model_class in self.all_model_classes:
if not model_class._supports_flash_attn_2:
return

4
tests/models/distilbert/test_modeling_distilbert.py
View File

@ -301,7 +301,7 @@ class DistilBertModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCa
@require_torch_accelerator
@pytest.mark.flash_attn_test
@slow
def test_flash_attn_2_inference(self):
def test_flash_attn_2_inference_equivalence(self):
import torch
for model_class in self.all_model_classes:
@ -353,7 +353,7 @@ class DistilBertModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCa
@require_torch_accelerator
@pytest.mark.flash_attn_test
@slow
def test_flash_attn_2_inference_padding_right(self):
def test_flash_attn_2_inference_equivalence_right_padding(self):
import torch
for model_class in self.all_model_classes:

2
tests/models/gemma/test_modeling_gemma.py
View File

@ -462,7 +462,7 @@ class GemmaModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixi
@require_torch_gpu
@pytest.mark.flash_attn_test
@slow
def test_flash_attn_2_inference_padding_right(self):
def test_flash_attn_2_inference_equivalence_right_padding(self):
self.skipTest("Gemma flash attention does not support right padding")
@require_torch_sdpa

2
tests/models/mistral/test_modeling_mistral.py
View File

@ -466,7 +466,7 @@ class MistralModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMi
@require_torch_gpu
@pytest.mark.flash_attn_test
@slow
def test_flash_attn_2_inference_padding_right(self):
def test_flash_attn_2_inference_equivalence_right_padding(self):
self.skipTest("Mistral flash attention does not support right padding")

2
tests/models/mixtral/test_modeling_mixtral.py
View File

@ -465,7 +465,7 @@ class MixtralModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMi
@require_torch_gpu
@pytest.mark.flash_attn_test
@slow
def test_flash_attn_2_inference_padding_right(self):
def test_flash_attn_2_inference_equivalence_right_padding(self):
self.skipTest("Mixtral flash attention does not support right padding")
# Ignore copy

2
tests/models/qwen2/test_modeling_qwen2.py
View File

@ -477,7 +477,7 @@ class Qwen2ModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixi
@require_torch_gpu
@pytest.mark.flash_attn_test
@slow
def test_flash_attn_2_inference_padding_right(self):
def test_flash_attn_2_inference_equivalence_right_padding(self):
self.skipTest("Qwen2 flash attention does not support right padding")

2
tests/models/starcoder2/test_modeling_starcoder2.py
View File

@ -461,7 +461,7 @@ class Starcoder2ModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTeste
@require_torch_gpu
@pytest.mark.flash_attn_test
@slow
def test_flash_attn_2_inference_padding_right(self):
def test_flash_attn_2_inference_equivalence_right_padding(self):
self.skipTest("Starcoder2 flash attention does not support right padding")

4
tests/models/whisper/test_modeling_whisper.py
View File

@ -888,7 +888,7 @@ class WhisperModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMi
@require_torch_gpu
@pytest.mark.flash_attn_test
@slow
def test_flash_attn_2_inference(self):
def test_flash_attn_2_inference_equivalence(self):
import torch
for model_class in self.all_model_classes:
@ -934,7 +934,7 @@ class WhisperModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMi
@require_torch_gpu
@pytest.mark.flash_attn_test
@slow
def test_flash_attn_2_inference_padding_right(self):
def test_flash_attn_2_inference_equivalence_right_padding(self):
import torch
for model_class in self.all_model_classes:

12
tests/test_modeling_common.py
View File

@ -3245,7 +3245,7 @@ class ModelTesterMixin:
@require_torch_gpu
@mark.flash_attn_test
@slow
def test_flash_attn_2_inference(self):
def test_flash_attn_2_inference_equivalence(self):
for model_class in self.all_model_classes:
if not model_class._supports_flash_attn_2:
self.skipTest(f"{model_class.__name__} does not support Flash Attention 2")
@ -3260,9 +3260,7 @@ class ModelTesterMixin:
)
model_fa.to(torch_device)
model = model_class.from_pretrained(
tmpdirname, torch_dtype=torch.bfloat16, attn_implementation="flash_attention_2"
)
model = model_class.from_pretrained(tmpdirname, torch_dtype=torch.bfloat16)
model.to(torch_device)
dummy_input = inputs_dict[model.main_input_name][:1]
@ -3340,7 +3338,7 @@ class ModelTesterMixin:
@require_torch_gpu
@mark.flash_attn_test
@slow
def test_flash_attn_2_inference_padding_right(self):
def test_flash_attn_2_inference_equivalence_right_padding(self):
for model_class in self.all_model_classes:
if not model_class._supports_flash_attn_2:
self.skipTest(f"{model_class.__name__} does not support Flash Attention 2")
@ -3355,9 +3353,7 @@ class ModelTesterMixin:
)
model_fa.to(torch_device)
model = model_class.from_pretrained(
tmpdirname, torch_dtype=torch.bfloat16, attn_implementation="flash_attention_2"
)
model = model_class.from_pretrained(tmpdirname, torch_dtype=torch.bfloat16)
model.to(torch_device)
dummy_input = inputs_dict[model.main_input_name][:1]