Skip to content

Fix Apr 05 integration test regressions (cuda sm_86) #48170

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
ydshieh merged 8 commits into from
Aug 21, 2026
+30 −5
Merged

Fix Apr 05 integration test regressions (cuda sm_86) #48170

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
18b5264
[MusicgenMelody] Add ("cuda", 8) expected values for 8 integration te…
ydshieh Aug 21, 2026
d4dd240
[StableLm] Add ("cuda", 8) expected values for 2 integration tests (c…
ydshieh Aug 21, 2026
20ba36c
[Whisper] Add ("cuda", 8) expected transcription for test_speculative…
ydshieh Aug 21, 2026
f3b4084
[EomtDinov3] Update stale ("cuda", 8) expected values in test_inferen…
ydshieh Aug 21, 2026
4135798
[GLM] Update stale ("cuda", 8) expected text in test_model_9b_eager (…
ydshieh Aug 21, 2026
48bdb52
[Dia] Add ("cuda", 8) expected token tensors in test_dia_model_integr…
ydshieh Aug 21, 2026
0f0a19a
[RAG] Add ("cuda", 8) expected outputs for token/sequence beam search…
ydshieh Aug 21, 2026
25e5204
Revert dia, eomt_dinov3, musicgen_melody, rag, whisper — unreliable A…
ydshieh Aug 21, 2026
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
2 changes: 1 addition & 1 deletion tests/models/glm/test_modeling_glm.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -107,7 +107,7 @@ def test_model_9b_eager(self):
],
("cuda", 8): [
'Hello I am doing a project on the history of the internetSolution:\n\nStep 1: Introduction\nThe history of the',
'Hi today I am going to show you how to make a simple and easy to make a DIY paper lantern.',
'Hi today I am going to show you how to make a simple and easy to make a DIY paper flower.',
],
("xpu", 5): [
"Hello I am doing a project on the history of the internetSolution:\n\nStep 1: Introduction\nThe history of the",
Expand Down
33 changes: 29 additions & 4 deletions tests/models/stablelm/test_modeling_stablelm.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -19,6 +19,7 @@

from transformers import BitsAndBytesConfig, is_torch_available
from transformers.testing_utils import (
Expectations,
require_bitsandbytes,
require_flash_attn,
require_torch,
Expand Down Expand Up @@ -61,11 +62,23 @@ def test_model_stablelm_3b_4e1t_logits(self):
output = model(**input_ids).logits.float()

# Expected mean on dim = -1
EXPECTED_MEAN = torch.tensor([[2.7146, 2.4245, 1.5616, 1.4424, 2.6790]]).to(torch_device)
expectations_mean = Expectations(
{
(None, None): [[2.7146, 2.4245, 1.5616, 1.4424, 2.6790]],
("cuda", 8): [[2.7304, 2.4242, 1.5718, 1.4360, 2.6792]],
}
) # fmt: skip
EXPECTED_MEAN = torch.tensor(expectations_mean.get_expectation()).to(torch_device)
torch.testing.assert_close(output.mean(dim=-1), EXPECTED_MEAN, rtol=1e-4, atol=1e-4)

# Expected logits sliced from [0, 0, 0:30]
EXPECTED_SLICE = torch.tensor([7.1030, -1.4195, 9.9206, 7.7008, 4.9891, 4.2169, 5.5426, 3.7878, 6.7593, 5.7360, 8.4691, 5.5448, 5.0544, 10.4129, 8.5573, 13.0405, 7.3265, 3.5868, 6.1106, 5.9406, 5.6376, 5.7490, 5.4850, 4.8124, 5.1991, 4.6419, 4.5719, 9.9588, 6.7222, 4.5070]).to(torch_device) # fmt: skip
expectations_slice = Expectations(
{
(None, None): [7.1030, -1.4195, 9.9206, 7.7008, 4.9891, 4.2169, 5.5426, 3.7878, 6.7593, 5.7360, 8.4691, 5.5448, 5.0544, 10.4129, 8.5573, 13.0405, 7.3265, 3.5868, 6.1106, 5.9406, 5.6376, 5.7490, 5.4850, 4.8124, 5.1991, 4.6419, 4.5719, 9.9588, 6.7222, 4.5070],
("cuda", 8): [7.1563, -1.4141, 9.8125, 7.7813, 4.9688, 4.3438, 5.2188, 3.3281, 6.6563, 5.9375, 8.3750, 5.3125, 4.7188, 10.2500, 8.6250, 13.0000, 7.2500, 3.4063, 5.8125, 5.6875, 5.3750, 5.4688, 5.2813, 4.5625, 4.9688, 4.4063, 4.3125, 10.0625, 6.7813, 4.5625],
}
) # fmt: skip
EXPECTED_SLICE = torch.tensor(expectations_slice.get_expectation()).to(torch_device)
torch.testing.assert_close(output[0, 0, :30], EXPECTED_SLICE, rtol=1e-4, atol=1e-4)

@slow
Expand Down Expand Up @@ -94,11 +107,23 @@ def test_model_tiny_random_stablelm_2_logits(self):
output = model(**input_ids).logits.float()

# Expected mean on dim = -1
EXPECTED_MEAN = torch.tensor([[-2.7196, -3.6099, -2.6877, -3.1973, -3.9344]]).to(torch_device)
expectations_mean = Expectations(
{
(None, None): [[-2.7196, -3.6099, -2.6877, -3.1973, -3.9344]],
("cuda", 8): [[-2.7165, -3.6102, -2.6881, -3.1981, -3.9231]],
}
) # fmt: skip
EXPECTED_MEAN = torch.tensor(expectations_mean.get_expectation()).to(torch_device)
torch.testing.assert_close(output.mean(dim=-1), EXPECTED_MEAN, rtol=1e-4, atol=1e-4)

# Expected logits sliced from [0, 0, 0:30]
EXPECTED_SLICE = torch.tensor([2.8364, 5.3811, 5.1659, 7.5485, 4.3219, 6.3315, 1.3967, 6.9147, 3.9679, 6.4786, 5.9176, 3.3067, 5.2917, 0.1485, 3.9630, 7.9947,10.6727, 9.6757, 8.8772, 8.3527, 7.8445, 6.6025, 5.5786, 7.0985,6.1369, 3.4259, 1.9397, 4.6157, 4.8105, 3.1768]).to(torch_device) # fmt: skip
expectations_slice = Expectations(
{
(None, None): [2.8364, 5.3811, 5.1659, 7.5485, 4.3219, 6.3315, 1.3967, 6.9147, 3.9679, 6.4786, 5.9176, 3.3067, 5.2917, 0.1485, 3.9630, 7.9947, 10.6727, 9.6757, 8.8772, 8.3527, 7.8445, 6.6025, 5.5786, 7.0985, 6.1369, 3.4259, 1.9397, 4.6157, 4.8105, 3.1768],
("cuda", 8): [2.8438, 5.3750, 5.1563, 7.5625, 4.2813, 6.3125, 1.3750, 6.9063, 3.9375, 6.5000, 5.9063, 3.3125, 5.2813, 0.1240, 3.9531, 7.9688, 10.6875, 9.6875, 8.8750, 8.3750, 7.8438, 6.5938, 5.5625, 7.0938, 6.1250, 3.4219, 1.9375, 4.5938, 4.7813, 3.1719],
}
) # fmt: skip
EXPECTED_SLICE = torch.tensor(expectations_slice.get_expectation()).to(torch_device)
torch.testing.assert_close(output[0, 0, :30], EXPECTED_SLICE, rtol=1e-4, atol=1e-4)

@slow
Expand Down
Loading