use torch.testing.assertclose instead to get more details about error in cis (#35659)

* use torch.testing.assertclose instead to get more details about error in cis

* fix

* style

* test_all

* revert for I bert

* fixes and updates

* more image processing fixes

* more image processors

* fix mamba and co

* style

* less strick

* ok I won't be strict

* skip and be done

* up
This commit is contained in:
Arthur
2025-01-24 16:55:28 +01:00
committed by GitHub
parent 72d1a4cd53
commit b912f5ee43
255 changed files with 1048 additions and 969 deletions

View File

@@ -342,7 +342,7 @@ class GraniteMoeModelTest(ModelTesterMixin, GenerationTesterMixin, unittest.Test
# Dynamic scaling does not change the RoPE embeddings until it receives an input longer than the original
# maximum sequence length, so the outputs for the short input should match.
if scaling_type == "dynamic":
self.assertTrue(torch.allclose(original_short_output, scaled_short_output, atol=1e-5))
torch.testing.assert_close(original_short_output, scaled_short_output, rtol=1e-5, atol=1e-5)
else:
self.assertFalse(torch.allclose(original_short_output, scaled_short_output, atol=1e-5))
@@ -441,9 +441,7 @@ class GraniteMoeIntegrationTest(unittest.TestCase):
# Expected mean on dim = -1
EXPECTED_MEAN = torch.tensor([[-2.2122, -1.6632, -2.9269, -2.3344, -2.0143, -3.0146, -2.6839, -2.5610]])
self.assertTrue(
torch.allclose(EXPECTED_MEAN.to(torch_device), out.logits.float().mean(-1), atol=1e-2, rtol=1e-2)
)
torch.testing.assert_close(EXPECTED_MEAN.to(torch_device), out.logits.float().mean(-1), rtol=1e-2, atol=1e-2)
# slicing logits[0, 0, 0:15]
EXPECTED_SLICE = torch.tensor([[4.8785, -2.2890, -2.2892, -2.2885, -2.2890, -3.5007, -2.2897, -2.2892,