Update diarization and WavLM tolerances (#14902)

tests/test_modeling_unispeech_sat.py

@@ -889,7 +889,8 @@ class UniSpeechSatModelIntegrationTest(unittest.TestCase):
         )
         self.assertEqual(labels[0, :, 0].sum(), 270)
         self.assertEqual(labels[0, :, 1].sum(), 647)
-        self.assertTrue(torch.allclose(outputs.logits[:, :4], expected_logits, atol=1e-3))
+        # TODO: update the tolerance after the CI moves to torch 1.10
+        self.assertTrue(torch.allclose(outputs.logits[:, :4], expected_logits, atol=1e-2))
 
     def test_inference_speaker_verification(self):
         model = UniSpeechSatForXVector.from_pretrained("microsoft/unispeech-sat-base-plus-sv").to(torch_device)
@@ -913,4 +914,5 @@ class UniSpeechSatModelIntegrationTest(unittest.TestCase):
         # id10002 vs id10004
         self.assertAlmostEqual(cosine_sim(embeddings[2], embeddings[3]).item(), 0.5616, 3)
 
-        self.assertAlmostEqual(outputs.loss.item(), 18.5925, 3)
+        # TODO: update the tolerance after the CI moves to torch 1.10
+        self.assertAlmostEqual(outputs.loss.item(), 18.5925, 2)

tests/test_modeling_wav2vec2.py

@@ -1480,7 +1480,8 @@ class Wav2Vec2ModelIntegrationTest(unittest.TestCase):
         )
         self.assertEqual(labels[0, :, 0].sum(), 555)
         self.assertEqual(labels[0, :, 1].sum(), 299)
-        self.assertTrue(torch.allclose(outputs.logits[:, :4], expected_logits, atol=1e-3))
+        # TODO: update the tolerance after the CI moves to torch 1.10
+        self.assertTrue(torch.allclose(outputs.logits[:, :4], expected_logits, atol=1e-2))
 
     def test_inference_speaker_verification(self):
         model = Wav2Vec2ForXVector.from_pretrained("anton-l/wav2vec2-base-superb-sv").to(torch_device)
@@ -1504,4 +1505,5 @@ class Wav2Vec2ModelIntegrationTest(unittest.TestCase):
         # id10002 vs id10004
         self.assertAlmostEqual(cosine_sim(embeddings[2], embeddings[3]).numpy(), 0.7594, 3)
 
-        self.assertAlmostEqual(outputs.loss.item(), 17.7963, 3)
+        # TODO: update the tolerance after the CI moves to torch 1.10
+        self.assertAlmostEqual(outputs.loss.item(), 17.7963, 2)

tests/test_modeling_wavlm.py

@@ -496,7 +496,8 @@ class WavLMModelIntegrationTest(unittest.TestCase):
         EXPECTED_HIDDEN_STATES_SLICE = torch.tensor(
             [[[0.0577, 0.1161], [0.0579, 0.1165]], [[0.0199, 0.1237], [0.0059, 0.0605]]]
         )
-        self.assertTrue(torch.allclose(hidden_states_slice, EXPECTED_HIDDEN_STATES_SLICE, rtol=1e-2))
+        # TODO: update the tolerance after the CI moves to torch 1.10
+        self.assertTrue(torch.allclose(hidden_states_slice, EXPECTED_HIDDEN_STATES_SLICE, atol=1e-2))
 
     def test_inference_large(self):
         model = WavLMModel.from_pretrained("microsoft/wavlm-large").to(torch_device)
@@ -546,7 +547,8 @@ class WavLMModelIntegrationTest(unittest.TestCase):
         )
         self.assertEqual(labels[0, :, 0].sum(), 258)
         self.assertEqual(labels[0, :, 1].sum(), 647)
-        self.assertTrue(torch.allclose(outputs.logits[:, :4], expected_logits, atol=1e-3))
+        # TODO: update the tolerance after the CI moves to torch 1.10
+        self.assertTrue(torch.allclose(outputs.logits[:, :4], expected_logits, atol=1e-2))
 
     def test_inference_speaker_verification(self):
         model = WavLMForXVector.from_pretrained("microsoft/wavlm-base-plus-sv").to(torch_device)
@@ -570,4 +572,5 @@ class WavLMModelIntegrationTest(unittest.TestCase):
         # id10002 vs id10004
         self.assertAlmostEqual(cosine_sim(embeddings[2], embeddings[3]).item(), 0.4780, 3)
 
-        self.assertAlmostEqual(outputs.loss.item(), 18.4154, 3)
+        # TODO: update the tolerance after the CI moves to torch 1.10
+        self.assertAlmostEqual(outputs.loss.item(), 18.4154, 2)