first commit

tests/models/camembert/test_modeling_camembert.py

@@ -0,0 +1,81 @@
+# Copyright 2020 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import unittest
+
+from transformers import is_torch_available
+from transformers.testing_utils import (
+    require_sentencepiece,
+    require_tokenizers,
+    require_torch,
+    slow,
+    torch_device,
+)
+
+
+if is_torch_available():
+    import torch
+
+    from transformers import CamembertModel
+
+
+@require_torch
+@require_sentencepiece
+@require_tokenizers
+class CamembertModelIntegrationTest(unittest.TestCase):
+    @slow
+    def test_output_embeds_base_model(self):
+        model = CamembertModel.from_pretrained("almanach/camembert-base", attn_implementation="eager")
+        model.to(torch_device)
+
+        input_ids = torch.tensor(
+            [[5, 121, 11, 660, 16, 730, 25543, 110, 83, 6]],
+            device=torch_device,
+            dtype=torch.long,
+        )  # J'aime le camembert !
+        with torch.no_grad():
+            output = model(input_ids)["last_hidden_state"]
+        expected_shape = torch.Size((1, 10, 768))
+        self.assertEqual(output.shape, expected_shape)
+        # compare the actual values for a slice.
+        expected_slice = torch.tensor(
+            [[[-0.0254, 0.0235, 0.1027], [0.0606, -0.1811, -0.0418], [-0.1561, -0.1127, 0.2687]]],
+            device=torch_device,
+            dtype=torch.float,
+        )
+        # camembert = torch.hub.load('pytorch/fairseq', 'camembert.v0')
+        # camembert.eval()
+        # expected_slice = roberta.model.forward(input_ids)[0][:, :3, :3].detach()
+
+        torch.testing.assert_close(output[:, :3, :3], expected_slice, rtol=1e-4, atol=1e-4)
+
+    @slow
+    def test_output_embeds_base_model_sdpa(self):
+        input_ids = torch.tensor(
+            [[5, 121, 11, 660, 16, 730, 25543, 110, 83, 6]],
+            device=torch_device,
+            dtype=torch.long,
+        )  # J'aime le camembert !
+
+        expected_slice = torch.tensor(
+            [[[-0.0254, 0.0235, 0.1027], [0.0606, -0.1811, -0.0418], [-0.1561, -0.1127, 0.2687]]],
+            device=torch_device,
+            dtype=torch.float,
+        )
+
+        model = CamembertModel.from_pretrained("almanach/camembert-base", attn_implementation="sdpa").to(torch_device)
+        with torch.no_grad():
+            output = model(input_ids)["last_hidden_state"].detach()
+
+        torch.testing.assert_close(output[:, :3, :3], expected_slice, rtol=1e-4, atol=1e-4)

tests/models/camembert/test_tokenization_camembert.py

@@ -0,0 +1,17 @@
+import unittest
+
+from transformers.models.camembert.tokenization_camembert import CamembertTokenizer
+from transformers.testing_utils import require_tokenizers
+
+from ...test_tokenization_common import TokenizerTesterMixin
+
+
+@require_tokenizers
+class CamembertTokenizationTest(TokenizerTesterMixin, unittest.TestCase):
+    from_pretrained_id = ["almanach/camembert-base"]
+    tokenizer_class = CamembertTokenizer
+
+    integration_expected_tokens = ['▁This', '▁is', '▁a', '▁test', '▁', '😊', '▁I', '▁was', '▁', 'born', '▁in', '▁9', '2000', ',', '▁and', '▁this', '▁is', '▁fal', 'sé', '.', '▁', '生活的真谛是', '▁Hi', '▁Hello', '▁Hi', '▁Hello', '▁Hello', '<s>', '▁hi', '<s>', '▁the', 're', '▁The', '▁', 'follow', 'ing', '▁string', '▁s', 'h', 'ould', '▁be', '▁pro', 'per', 'ly', '▁en', 'code', 'd', ':', '▁Hello', '.', '▁But', '▁i', 'rd', '▁and', '▁', 'ปี', '▁i', 'rd', '▁', 'ด', '▁Hey', '▁h', 'ow', '▁are', '▁you', '▁do', 'ing']  # fmt: skip
+    integration_expected_token_ids = [17526, 2856, 33, 2006, 21, 3, 551, 15760, 21, 24900, 378, 419, 13233, 7, 1168, 9098, 2856, 19289, 5100, 9, 21, 3, 5108, 9774, 5108, 9774, 9774, 5, 7874, 5, 808, 346, 908, 21, 31189, 402, 20468, 52, 133, 19306, 2446, 909, 1399, 1107, 22, 14420, 204, 92, 9774, 9, 10503, 1723, 6682, 1168, 21, 3, 1723, 6682, 21, 3, 20128, 616, 3168, 9581, 4835, 7503, 402]  # fmt: skip
+    expected_tokens_from_ids = ['▁This', '▁is', '▁a', '▁test', '▁', '<unk>', '▁I', '▁was', '▁', 'born', '▁in', '▁9', '2000', ',', '▁and', '▁this', '▁is', '▁fal', 'sé', '.', '▁', '<unk>', '▁Hi', '▁Hello', '▁Hi', '▁Hello', '▁Hello', '<s>', '▁hi', '<s>', '▁the', 're', '▁The', '▁', 'follow', 'ing', '▁string', '▁s', 'h', 'ould', '▁be', '▁pro', 'per', 'ly', '▁en', 'code', 'd', ':', '▁Hello', '.', '▁But', '▁i', 'rd', '▁and', '▁', '<unk>', '▁i', 'rd', '▁', '<unk>', '▁Hey', '▁h', 'ow', '▁are', '▁you', '▁do', 'ing']  # fmt: skip
+    integration_expected_decoded_text = "This is a test <unk> I was born in 92000, and this is falsé. <unk> Hi Hello Hi Hello Hello<s> hi<s> there The following string should be properly encoded: Hello. But ird and <unk> ird <unk> Hey how are you doing"