Skip to content

Add AMD test expectations to DETR model #39539

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 4 commits into from
Jul 22, 2025
Merged

Add AMD test expectations to DETR model #39539

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
b139531
Add AMD test expectations to DETR model
ahadnagy Jul 20, 2025
d845be7
Fix baseline expectation
ahadnagy Jul 20, 2025
8aca6fb
Address review comments
ahadnagy Jul 21, 2025
e74c37d
Make formatting a bit more consistent
ahadnagy Jul 22, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
138 changes: 103 additions & 35 deletions tests/models/detr/test_modeling_detr.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,7 +18,7 @@
import unittest

from transformers import DetrConfig, ResNetConfig, is_torch_available, is_vision_available
from transformers.testing_utils import require_timm, require_torch, require_vision, slow, torch_device
from transformers.testing_utils import Expectations, require_timm, require_torch, require_vision, slow, torch_device
from transformers.utils import cached_property

from ...test_configuration_common import ConfigTester
Expand Down Expand Up @@ -585,13 +585,23 @@ def test_inference_no_head(self):

expected_shape = torch.Size((1, 100, 256))
assert outputs.last_hidden_state.shape == expected_shape
expected_slice = torch.tensor(
[
[0.0622, -0.5142, -0.4034],
[-0.7628, -0.4935, -1.7153],
[-0.4751, -0.6386, -0.7818],
]
).to(torch_device)
expected_slices = Expectations(
{
(None, None):
[
[0.0622, -0.5142, -0.4034],
[-0.7628, -0.4935, -1.7153],
[-0.4751, -0.6386, -0.7818],
],
("rocm", (9, 5)):
[
[ 0.0616, -0.5146, -0.4032],
[-0.7629, -0.4934, -1.7153],
[-0.4768, -0.6403, -0.7826],
],
}
) # fmt: skip
expected_slice = torch.tensor(expected_slices.get_expectation(), device=torch_device)
torch.testing.assert_close(outputs.last_hidden_state[0, :3, :3], expected_slice, rtol=2e-4, atol=2e-4)

def test_inference_object_detection_head(self):
Expand All @@ -609,13 +619,23 @@ def test_inference_object_detection_head(self):
# verify outputs
expected_shape_logits = torch.Size((1, model.config.num_queries, model.config.num_labels + 1))
self.assertEqual(outputs.logits.shape, expected_shape_logits)
expected_slice_logits = torch.tensor(
[
[-19.1211, -0.0881, -11.0188],
[-17.3641, -1.8045, -14.0229],
[-20.0415, -0.5833, -11.1005],
]
).to(torch_device)
expected_slices = Expectations(
{
(None, None):
[
[-19.1211, -0.0881, -11.0188],
[-17.3641, -1.8045, -14.0229],
[-20.0415, -0.5833, -11.1005],
],
("rocm", (9, 5)):
[
[-19.1194, -0.0893, -11.0154],
[-17.3640, -1.8035, -14.0219],
[-20.0461, -0.5837, -11.1060],
],
}
) # fmt: skip
expected_slice_logits = torch.tensor(expected_slices.get_expectation(), device=torch_device)
torch.testing.assert_close(outputs.logits[0, :3, :3], expected_slice_logits, rtol=2e-4, atol=2e-4)

expected_shape_boxes = torch.Size((1, model.config.num_queries, 4))
Expand Down Expand Up @@ -657,27 +677,65 @@ def test_inference_panoptic_segmentation_head(self):
# verify outputs
expected_shape_logits = torch.Size((1, model.config.num_queries, model.config.num_labels + 1))
self.assertEqual(outputs.logits.shape, expected_shape_logits)
expected_slice_logits = torch.tensor(
[
[-18.1523, -1.7592, -13.5019],
[-16.8866, -1.4139, -14.1025],
[-17.5735, -2.5090, -11.8666],
]
).to(torch_device)
expected_slices = Expectations(
{
(None, None):
[
[-18.1523, -1.7592, -13.5019],
[-16.8866, -1.4139, -14.1025],
[-17.5735, -2.5090, -11.8666],
],
("rocm", (9, 5)):
[
[-18.1565, -1.7568, -13.5029],
[-16.8888, -1.4138, -14.1028],
[-17.5709, -2.5080, -11.8654],
],
}
) # fmt: skip
expected_slice_logits = torch.tensor(expected_slices.get_expectation(), device=torch_device)
torch.testing.assert_close(outputs.logits[0, :3, :3], expected_slice_logits, rtol=2e-4, atol=2e-4)

expected_shape_boxes = torch.Size((1, model.config.num_queries, 4))
self.assertEqual(outputs.pred_boxes.shape, expected_shape_boxes)
expected_slice_boxes = torch.tensor(
[[0.5344, 0.1790, 0.9284], [0.4421, 0.0571, 0.0875], [0.6632, 0.6886, 0.1015]]
).to(torch_device)
expected_slices = Expectations(
{
(None, None):
[
[0.5344, 0.1790, 0.9284],
[0.4421, 0.0571, 0.0875],
[0.6632, 0.6886, 0.1015]
],
("rocm", (9, 5)):
[
[0.5344, 0.1789, 0.9285],
[0.4420, 0.0572, 0.0875],
[0.6630, 0.6887, 0.1017],
],
}
) # fmt: skip
expected_slice_boxes = torch.tensor(expected_slices.get_expectation(), device=torch_device)
torch.testing.assert_close(outputs.pred_boxes[0, :3, :3], expected_slice_boxes, rtol=2e-4, atol=2e-4)

expected_shape_masks = torch.Size((1, model.config.num_queries, 200, 267))
self.assertEqual(outputs.pred_masks.shape, expected_shape_masks)
expected_slice_masks = torch.tensor(
[[-7.8408, -11.0104, -12.1279], [-12.0299, -16.6498, -17.9806], [-14.8995, -19.9940, -20.5646]]
).to(torch_device)
expected_slices = Expectations(
{
(None, None):
[
[-7.8408, -11.0104, -12.1279],
[-12.0299, -16.6498, -17.9806],
[-14.8995, -19.9940, -20.5646],
],
("rocm", (9, 5)):
[
[ -7.7558, -10.8789, -11.9798],
[-11.8882, -16.4330, -17.7452],
[-14.7317, -19.7384, -20.3005],
],
}
) # fmt: skip
expected_slice_masks = torch.tensor(expected_slices.get_expectation(), device=torch_device)
torch.testing.assert_close(outputs.pred_masks[0, 0, :3, :3], expected_slice_masks, rtol=2e-3, atol=2e-3)

# verify postprocessing
Expand Down Expand Up @@ -731,11 +789,21 @@ def test_inference_no_head(self):

expected_shape = torch.Size((1, 100, 256))
assert outputs.last_hidden_state.shape == expected_shape
expected_slice = torch.tensor(
[
[0.0622, -0.5142, -0.4034],
[-0.7628, -0.4935, -1.7153],
[-0.4751, -0.6386, -0.7818],
]
).to(torch_device)
expected_slices = Expectations(
{
(None, None):
[
[0.0622, -0.5142, -0.4034],
[-0.7628, -0.4935, -1.7153],
[-0.4751, -0.6386, -0.7818],
],
("rocm", (9, 5)):
[
[ 0.0616, -0.5146, -0.4032],
[-0.7629, -0.4934, -1.7153],
[-0.4768, -0.6403, -0.7826]
],
}
) # fmt: skip
expected_slice = torch.tensor(expected_slices.get_expectation(), device=torch_device)
torch.testing.assert_close(outputs.last_hidden_state[0, :3, :3], expected_slice, rtol=1e-4, atol=1e-4)