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

fix(DPT,Depth-Anything) torch.export (#34103)

Browse Source 
* Fix torch.export issue in dpt based models

Signed-off-by: Phillip Kuznetsov <philkuz@gimletlabs.ai>

* Simplify the if statements

Signed-off-by: Phillip Kuznetsov <philkuz@gimletlabs.ai>

* Move activation definitions of zoe_depth to init()

Signed-off-by: Phillip Kuznetsov <philkuz@gimletlabs.ai>

* Add test_export for dpt and zoedepth

Signed-off-by: Phillip Kuznetsov <philkuz@gimletlabs.ai>

* add depth anything

Signed-off-by: Phillip Kuznetsov <philkuz@gimletlabs.ai>

* Remove zoedepth non-automated zoedepth changes and zoedepth test

Signed-off-by: Phillip Kuznetsov <philkuz@gimletlabs.ai>

* [run_slow] dpt, depth_anything, zoedepth

Signed-off-by: Phillip Kuznetsov <philkuz@gimletlabs.ai>

---------

Signed-off-by: Phillip Kuznetsov <philkuz@gimletlabs.ai>
This commit is contained in:
Phillip Kuznetsov
2024-11-20 02:31:21 -08:00
committed by GitHub
parent 9d16441e4f
commit 8cadf76e1c
5 changed files with 72 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files

28
tests/models/depth_anything/test_modeling_depth_anything.py
View File

@@ -18,6 +18,7 @@ import unittest
from transformers import DepthAnythingConfig, Dinov2Config
from transformers.file_utils import is_torch_available, is_vision_available
from transformers.pytorch_utils import is_torch_greater_or_equal_than_2_4
from transformers.testing_utils import require_torch, require_vision, slow, torch_device
from ...test_configuration_common import ConfigTester
@@ -290,3 +291,30 @@ class DepthAnythingModelIntegrationTest(unittest.TestCase):
).to(torch_device)
self.assertTrue(torch.allclose(predicted_depth[0, :3, :3], expected_slice, atol=1e-4))
def test_export(self):
for strict in [True, False]:
with self.subTest(strict=strict):
if not is_torch_greater_or_equal_than_2_4:
self.skipTest(reason="This test requires torch >= 2.4 to run.")
model = (
DepthAnythingForDepthEstimation.from_pretrained("LiheYoung/depth-anything-small-hf")
.to(torch_device)
.eval()
)
image_processor = DPTImageProcessor.from_pretrained("LiheYoung/depth-anything-small-hf")
image = prepare_img()
inputs = image_processor(images=image, return_tensors="pt").to(torch_device)
exported_program = torch.export.export(
model,
args=(inputs["pixel_values"],),
strict=strict,
)
with torch.no_grad():
eager_outputs = model(**inputs)
exported_outputs = exported_program.module().forward(inputs["pixel_values"])
self.assertEqual(eager_outputs.predicted_depth.shape, exported_outputs.predicted_depth.shape)
self.assertTrue(
torch.allclose(eager_outputs.predicted_depth, exported_outputs.predicted_depth, atol=1e-4)
)

22
tests/models/dpt/test_modeling_dpt.py
View File

@@ -18,6 +18,7 @@ import unittest
from transformers import DPTConfig
from transformers.file_utils import is_torch_available, is_vision_available
from transformers.pytorch_utils import is_torch_greater_or_equal_than_2_4
from transformers.testing_utils import require_torch, require_vision, slow, torch_device
from ...test_configuration_common import ConfigTester
@@ -410,3 +411,24 @@ class DPTModelIntegrationTest(unittest.TestCase):
).squeeze()
self.assertTrue(output_enlarged.shape == expected_shape)
self.assertTrue(torch.allclose(predicted_depth_l, output_enlarged, rtol=1e-3))
def test_export(self):
for strict in [True, False]:
with self.subTest(strict=strict):
if not is_torch_greater_or_equal_than_2_4:
self.skipTest(reason="This test requires torch >= 2.4 to run.")
model = DPTForSemanticSegmentation.from_pretrained("Intel/dpt-large-ade").to(torch_device).eval()
image_processor = DPTImageProcessor.from_pretrained("Intel/dpt-large-ade")
image = prepare_img()
inputs = image_processor(images=image, return_tensors="pt").to(torch_device)
exported_program = torch.export.export(
model,
args=(inputs["pixel_values"],),
strict=strict,
)
with torch.no_grad():
eager_outputs = model(**inputs)
exported_outputs = exported_program.module().forward(inputs["pixel_values"])
self.assertEqual(eager_outputs.logits.shape, exported_outputs.logits.shape)
self.assertTrue(torch.allclose(eager_outputs.logits, exported_outputs.logits, atol=1e-4))