[feat] Add FLAVA model (#16654)

* [WIP] Add FLAVA model

This PR aims to add [FLAVA](ihttps://arxiv.org/abs/2112.04482) model to the transformers repo.

Following checklist delineates the list of things to be done for this PR
to be complete:

[x] Flava init
[x] Flava base models
[x] Flava layers
[x] Flava Configs
[x] Flava encoders
[x] Flava pretraining models
[ ] Flava classification/retrieval models (To be added in a separate PR)
[x] Documentation updates 
[x] Imports updates 
[x] Argstring updates
[x] Flava pretrained checkpoints 
[x] Flava tests
[x] Flava processors 
[x] Sanity check
[x] Lint

src/transformers/utils/dummy_pt_objects.py

@@ -1787,6 +1787,58 @@ class FlaubertWithLMHeadModel(metaclass=DummyObject):
         requires_backends(self, ["torch"])
 
 
+FLAVA_PRETRAINED_MODEL_ARCHIVE_LIST = None
+
+
+class FlavaForPreTraining(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class FlavaImageCodebook(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class FlavaImageModel(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class FlavaModel(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class FlavaMultimodalModel(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class FlavaPreTrainedModel(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class FlavaTextModel(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
 FNET_PRETRAINED_MODEL_ARCHIVE_LIST = None
 
 

src/transformers/utils/dummy_vision_objects.py

@@ -59,6 +59,20 @@ class DPTFeatureExtractor(metaclass=DummyObject):
         requires_backends(self, ["vision"])
 
 
+class FlavaFeatureExtractor(metaclass=DummyObject):
+    _backends = ["vision"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["vision"])
+
+
+class FlavaProcessor(metaclass=DummyObject):
+    _backends = ["vision"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["vision"])
+
+
 class GLPNFeatureExtractor(metaclass=DummyObject):
     _backends = ["vision"]
 

tests/models/flava/test_feature_extraction_flava.py

+# coding=utf-8
+# Copyright 2022 Meta Platforms authors and HuggingFace Inc.
+#
+# 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 random
+import unittest
+
+import numpy as np
+
+from transformers.testing_utils import require_torch, require_vision
+from transformers.utils import is_torch_available, is_vision_available
+
+from ...test_feature_extraction_common import FeatureExtractionSavingTestMixin, prepare_image_inputs
+
+
+if is_torch_available():
+    import torch
+
+if is_vision_available():
+    from PIL import Image
+
+    from transformers import FlavaFeatureExtractor
+    from transformers.models.flava.feature_extraction_flava import (
+        FLAVA_CODEBOOK_MEAN,
+        FLAVA_CODEBOOK_STD,
+        FLAVA_IMAGE_MEAN,
+        FLAVA_IMAGE_STD,
+    )
+else:
+    FLAVA_IMAGE_MEAN = FLAVA_IMAGE_STD = FLAVA_CODEBOOK_MEAN = FLAVA_CODEBOOK_STD = None
+
+
+class FlavaFeatureExtractionTester(unittest.TestCase):
+    def __init__(
+        self,
+        parent,
+        batch_size=7,
+        num_channels=3,
+        min_resolution=30,
+        max_resolution=400,
+        do_resize=True,
+        size=224,
+        do_center_crop=True,
+        crop_size=224,
+        resample=None,
+        do_normalize=True,
+        image_mean=FLAVA_IMAGE_MEAN,
+        image_std=FLAVA_IMAGE_STD,
+        input_size_patches=14,
+        total_mask_patches=75,
+        mask_group_max_patches=None,
+        mask_group_min_patches=16,
+        mask_group_min_aspect_ratio=0.3,
+        mask_group_max_aspect_ratio=None,
+        codebook_do_resize=True,
+        codebook_size=112,
+        codebook_resample=None,
+        codebook_do_center_crop=True,
+        codebook_crop_size=112,
+        codebook_do_map_pixels=True,
+        codebook_do_normalize=True,
+        codebook_image_mean=FLAVA_CODEBOOK_MEAN,
+        codebook_image_std=FLAVA_CODEBOOK_STD,
+    ):
+        self.parent = parent
+        self.batch_size = batch_size
+        self.num_channels = num_channels
+        self.do_resize = do_resize
+        self.min_resolution = min_resolution
+        self.max_resolution = max_resolution
+        self.size = size
+        self.resample = resample if resample is not None else Image.BICUBIC
+        self.do_normalize = do_normalize
+        self.image_mean = image_mean
+        self.image_std = image_std
+        self.do_center_crop = do_center_crop
+        self.crop_size = crop_size
+
+        self.input_size_patches = input_size_patches
+        self.total_mask_patches = total_mask_patches
+        self.mask_group_max_patches = mask_group_max_patches
+        self.mask_group_min_patches = mask_group_min_patches
+        self.mask_group_min_aspect_ratio = mask_group_min_aspect_ratio
+        self.mask_group_max_aspect_ratio = mask_group_max_aspect_ratio
+
+        self.codebook_do_resize = codebook_do_resize
+        self.codebook_size = codebook_size
+        self.codebook_resample = codebook_resample if codebook_resample is not None else Image.LANCZOS
+        self.codebook_do_center_crop = codebook_do_center_crop
+        self.codebook_crop_size = codebook_crop_size
+        self.codebook_do_map_pixels = codebook_do_map_pixels
+        self.codebook_do_normalize = codebook_do_normalize
+        self.codebook_image_mean = codebook_image_mean
+        self.codebook_image_std = codebook_image_std
+
+    def prepare_feat_extract_dict(self):
+        return {
+            "image_mean": self.image_mean,
+            "image_std": self.image_std,
+            "do_normalize": self.do_normalize,
+            "do_resize": self.do_resize,
+            "size": self.size,
+            "resample": self.resample,
+            "do_center_crop": self.do_center_crop,
+            "crop_size": self.crop_size,
+            "input_size_patches": self.input_size_patches,
+            "total_mask_patches": self.total_mask_patches,
+            "mask_group_max_patches": self.mask_group_max_patches,
+            "mask_group_min_patches": self.mask_group_min_patches,
+            "mask_group_min_aspect_ratio": self.mask_group_min_aspect_ratio,
+            "mask_group_max_aspect_ratio": self.mask_group_min_aspect_ratio,
+            "codebook_do_resize": self.codebook_do_resize,
+            "codebook_size": self.codebook_size,
+            "codebook_resample": self.codebook_resample,
+            "codebook_do_center_crop": self.codebook_do_center_crop,
+            "codebook_crop_size": self.codebook_crop_size,
+            "codebook_do_map_pixels": self.codebook_do_map_pixels,
+            "codebook_do_normalize": self.codebook_do_normalize,
+            "codebook_image_mean": self.codebook_image_mean,
+            "codebook_image_std": self.codebook_image_std,
+        }
+
+    def get_expected_image_size(self):
+        return (self.size, self.size) if not isinstance(self.size, tuple) else self.size
+
+    def get_expected_mask_size(self):
+        return (
+            (self.input_size_patches, self.input_size_patches)
+            if not isinstance(self.input_size_patches, tuple)
+            else self.input_size_patches
+        )
+
+    def get_expected_codebook_image_size(self):
+        if not isinstance(self.codebook_size, tuple):
+            return (self.codebook_size, self.codebook_size)
+        else:
+            return self.codebook_size
+
+
+@require_torch
+@require_vision
+class FlavaFeatureExtractionTest(FeatureExtractionSavingTestMixin, unittest.TestCase):
+
+    feature_extraction_class = FlavaFeatureExtractor if is_vision_available() else None
+    maxDiff = None
+
+    def setUp(self):
+        self.feature_extract_tester = FlavaFeatureExtractionTester(self)
+
+    @property
+    def feat_extract_dict(self):
+        return self.feature_extract_tester.prepare_feat_extract_dict()
+
+    def test_feat_extract_properties(self):
+        feature_extractor = self.feature_extraction_class(**self.feat_extract_dict)
+        self.assertTrue(hasattr(feature_extractor, "image_mean"))
+        self.assertTrue(hasattr(feature_extractor, "image_std"))
+        self.assertTrue(hasattr(feature_extractor, "do_normalize"))
+        self.assertTrue(hasattr(feature_extractor, "do_resize"))
+        self.assertTrue(hasattr(feature_extractor, "resample"))
+        self.assertTrue(hasattr(feature_extractor, "crop_size"))
+        self.assertTrue(hasattr(feature_extractor, "do_center_crop"))
+        self.assertTrue(hasattr(feature_extractor, "masking_generator"))
+        self.assertTrue(hasattr(feature_extractor, "codebook_do_resize"))
+        self.assertTrue(hasattr(feature_extractor, "codebook_size"))
+        self.assertTrue(hasattr(feature_extractor, "codebook_resample"))
+        self.assertTrue(hasattr(feature_extractor, "codebook_do_center_crop"))
+        self.assertTrue(hasattr(feature_extractor, "codebook_crop_size"))
+        self.assertTrue(hasattr(feature_extractor, "codebook_do_map_pixels"))
+        self.assertTrue(hasattr(feature_extractor, "codebook_do_normalize"))
+        self.assertTrue(hasattr(feature_extractor, "codebook_image_mean"))
+        self.assertTrue(hasattr(feature_extractor, "codebook_image_std"))
+
+    def test_batch_feature(self):
+        pass
+
+    def test_call_pil(self):
+        # Initialize feature_extractor
+        feature_extractor = self.feature_extraction_class(**self.feat_extract_dict)
+        # create random PIL images
+        image_inputs = prepare_image_inputs(self.feature_extract_tester, equal_resolution=False)
+        for image in image_inputs:
+            self.assertIsInstance(image, Image.Image)
+
+        # Test not batched input
+        encoded_images = feature_extractor(image_inputs[0], return_tensors="pt")
+
+        # Test no bool masked pos
+        self.assertFalse("bool_masked_pos" in encoded_images)
+
+        expected_height, expected_width = self.feature_extract_tester.get_expected_image_size()
+
+        self.assertEqual(
+            encoded_images.pixel_values.shape,
+            (1, self.feature_extract_tester.num_channels, expected_height, expected_width),
+        )
+
+        # Test batched
+        encoded_images = feature_extractor(image_inputs, return_tensors="pt")
+        expected_height, expected_width = self.feature_extract_tester.get_expected_image_size()
+
+        # Test no bool masked pos
+        self.assertFalse("bool_masked_pos" in encoded_images)
+
+        self.assertEqual(
+            encoded_images.pixel_values.shape,
+            (
+                self.feature_extract_tester.batch_size,
+                self.feature_extract_tester.num_channels,
+                expected_height,
+                expected_width,
+            ),
+        )
+
+    def _test_call_framework(self, instance_class, prepare_kwargs):
+        # Initialize feature_extractor
+        feature_extractor = self.feature_extraction_class(**self.feat_extract_dict)
+        # create random tensors
+        image_inputs = prepare_image_inputs(self.feature_extract_tester, equal_resolution=False, **prepare_kwargs)
+        for image in image_inputs:
+            self.assertIsInstance(image, instance_class)
+
+        # Test not batched input
+        encoded_images = feature_extractor(image_inputs[0], return_tensors="pt")
+
+        expected_height, expected_width = self.feature_extract_tester.get_expected_image_size()
+        self.assertEqual(
+            encoded_images.pixel_values.shape,
+            (1, self.feature_extract_tester.num_channels, expected_height, expected_width),
+        )
+
+        encoded_images = feature_extractor(image_inputs, return_image_mask=True, return_tensors="pt")
+
+        expected_height, expected_width = self.feature_extract_tester.get_expected_image_size()
+        self.assertEqual(
+            encoded_images.pixel_values.shape,
+            (
+                self.feature_extract_tester.batch_size,
+                self.feature_extract_tester.num_channels,
+                expected_height,
+                expected_width,
+            ),
+        )
+
+        expected_height, expected_width = self.feature_extract_tester.get_expected_mask_size()
+        self.assertEqual(
+            encoded_images.bool_masked_pos.shape,
+            (
+                self.feature_extract_tester.batch_size,
+                expected_height,
+                expected_width,
+            ),
+        )
+
+        # Test batched
+        encoded_images = feature_extractor(image_inputs, return_tensors="pt").pixel_values
+
+        expected_height, expected_width = self.feature_extract_tester.get_expected_image_size()
+        self.assertEqual(
+            encoded_images.shape,
+            (
+                self.feature_extract_tester.batch_size,
+                self.feature_extract_tester.num_channels,
+                expected_height,
+                expected_width,
+            ),
+        )
+
+        # Test masking
+        encoded_images = feature_extractor(image_inputs, return_image_mask=True, return_tensors="pt")
+
+        expected_height, expected_width = self.feature_extract_tester.get_expected_image_size()
+        self.assertEqual(
+            encoded_images.pixel_values.shape,
+            (
+                self.feature_extract_tester.batch_size,
+                self.feature_extract_tester.num_channels,
+                expected_height,
+                expected_width,
+            ),
+        )
+
+        expected_height, expected_width = self.feature_extract_tester.get_expected_mask_size()
+        self.assertEqual(
+            encoded_images.bool_masked_pos.shape,
+            (
+                self.feature_extract_tester.batch_size,
+                expected_height,
+                expected_width,
+            ),
+        )
+
+    def test_call_numpy(self):
+        self._test_call_framework(np.ndarray, prepare_kwargs={"numpify": True})
+
+    def test_call_pytorch(self):
+        self._test_call_framework(torch.Tensor, prepare_kwargs={"torchify": True})
+
+    def test_masking(self):
+        # Initialize feature_extractor
+        random.seed(1234)
+        feature_extractor = self.feature_extraction_class(**self.feat_extract_dict)
+        image_inputs = prepare_image_inputs(self.feature_extract_tester, equal_resolution=False, torchify=True)
+
+        # Test not batched input
+        encoded_images = feature_extractor(image_inputs[0], return_image_mask=True, return_tensors="pt")
+        self.assertEqual(encoded_images.bool_masked_pos.sum().item(), 75)
+
+    def test_codebook_pixels(self):
+        # Initialize feature_extractor
+        feature_extractor = self.feature_extraction_class(**self.feat_extract_dict)
+        # create random PIL images
+        image_inputs = prepare_image_inputs(self.feature_extract_tester, equal_resolution=False)
+        for image in image_inputs:
+            self.assertIsInstance(image, Image.Image)
+
+        # Test not batched input
+        encoded_images = feature_extractor(image_inputs[0], return_codebook_pixels=True, return_tensors="pt")
+        expected_height, expected_width = self.feature_extract_tester.get_expected_codebook_image_size()
+        self.assertEqual(
+            encoded_images.codebook_pixel_values.shape,
+            (1, self.feature_extract_tester.num_channels, expected_height, expected_width),
+        )
+
+        # Test batched
+        encoded_images = feature_extractor(image_inputs, return_codebook_pixels=True, return_tensors="pt")
+        expected_height, expected_width = self.feature_extract_tester.get_expected_codebook_image_size()
+        self.assertEqual(
+            encoded_images.codebook_pixel_values.shape,
+            (
+                self.feature_extract_tester.batch_size,
+                self.feature_extract_tester.num_channels,
+                expected_height,
+                expected_width,
+            ),
+        )

tests/models/flava/test_processor_flava.py

+# Copyright 2022 Meta Platforms authors and 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 json
+import os
+import random
+import shutil
+import tempfile
+import unittest
+
+import numpy as np
+import pytest
+
+from transformers import BertTokenizer, BertTokenizerFast
+from transformers.models.bert.tokenization_bert import VOCAB_FILES_NAMES
+from transformers.testing_utils import require_vision
+from transformers.utils import FEATURE_EXTRACTOR_NAME, is_vision_available
+
+
+if is_vision_available():
+    from PIL import Image
+
+    from transformers import FlavaFeatureExtractor, FlavaProcessor
+    from transformers.models.flava.feature_extraction_flava import (
+        FLAVA_CODEBOOK_MEAN,
+        FLAVA_CODEBOOK_STD,
+        FLAVA_IMAGE_MEAN,
+        FLAVA_IMAGE_STD,
+    )
+
+
+@require_vision
+class FlavaProcessorTest(unittest.TestCase):
+    def setUp(self):
+        self.tmpdirname = tempfile.mkdtemp()
+
+        # fmt: off
+        vocab_tokens = ["[UNK]", "[CLS]", "[SEP]", "[PAD]", "[MASK]", "want", "##want", "##ed", "wa", "un", "runn", "##ing", ",", "low", "lowest"]
+        # fmt: on
+        self.vocab_file = os.path.join(self.tmpdirname, VOCAB_FILES_NAMES["vocab_file"])
+
+        with open(self.vocab_file, "w", encoding="utf-8") as fp:
+            fp.write("".join([x + "\n" for x in vocab_tokens]))
+
+        feature_extractor_map = {
+            "image_mean": FLAVA_IMAGE_MEAN,
+            "image_std": FLAVA_IMAGE_STD,
+            "do_normalize": True,
+            "do_resize": True,
+            "size": 224,
+            "do_center_crop": True,
+            "crop_size": 224,
+            "input_size_patches": 14,
+            "total_mask_patches": 75,
+            "mask_group_max_patches": None,
+            "mask_group_min_patches": 16,
+            "mask_group_min_aspect_ratio": 0.3,
+            "mask_group_max_aspect_ratio": None,
+            "codebook_do_resize": True,
+            "codebook_size": 112,
+            "codebook_resample": None,
+            "codebook_do_center_crop": True,
+            "codebook_crop_size": 112,
+            "codebook_do_map_pixels": True,
+            "codebook_do_normalize": True,
+            "codebook_image_mean": FLAVA_CODEBOOK_MEAN,
+            "codebook_image_std": FLAVA_CODEBOOK_STD,
+        }
+
+        self.feature_extractor_file = os.path.join(self.tmpdirname, FEATURE_EXTRACTOR_NAME)
+        with open(self.feature_extractor_file, "w", encoding="utf-8") as fp:
+            json.dump(feature_extractor_map, fp)
+
+    def get_tokenizer(self, **kwargs):
+        return BertTokenizer.from_pretrained(self.tmpdirname, **kwargs)
+
+    def get_rust_tokenizer(self, **kwargs):
+        return BertTokenizerFast.from_pretrained(self.tmpdirname, **kwargs)
+
+    def get_feature_extractor(self, **kwargs):
+        return FlavaFeatureExtractor.from_pretrained(self.tmpdirname, **kwargs)
+
+    def tearDown(self):
+        shutil.rmtree(self.tmpdirname)
+
+    def prepare_image_inputs(self):
+        """This function prepares a list of PIL images, or a list of numpy arrays if one specifies numpify=True,
+        or a list of PyTorch tensors if one specifies torchify=True.
+        """
+
+        image_inputs = [np.random.randint(255, size=(3, 30, 400), dtype=np.uint8)]
+
+        image_inputs = [Image.fromarray(np.moveaxis(x, 0, -1)) for x in image_inputs]
+
+        return image_inputs
+
+    def test_save_load_pretrained_default(self):
+        tokenizer_slow = self.get_tokenizer()
+        tokenizer_fast = self.get_rust_tokenizer()
+        feature_extractor = self.get_feature_extractor()
+
+        processor_slow = FlavaProcessor(tokenizer=tokenizer_slow, feature_extractor=feature_extractor)
+        processor_slow.save_pretrained(self.tmpdirname)
+        processor_slow = FlavaProcessor.from_pretrained(self.tmpdirname, use_fast=False)
+
+        processor_fast = FlavaProcessor(tokenizer=tokenizer_fast, feature_extractor=feature_extractor)
+        processor_fast.save_pretrained(self.tmpdirname)
+        processor_fast = FlavaProcessor.from_pretrained(self.tmpdirname)
+
+        self.assertEqual(processor_slow.tokenizer.get_vocab(), tokenizer_slow.get_vocab())
+        self.assertEqual(processor_fast.tokenizer.get_vocab(), tokenizer_fast.get_vocab())
+        self.assertEqual(tokenizer_slow.get_vocab(), tokenizer_fast.get_vocab())
+        self.assertIsInstance(processor_slow.tokenizer, BertTokenizer)
+        self.assertIsInstance(processor_fast.tokenizer, BertTokenizerFast)
+
+        self.assertEqual(processor_slow.feature_extractor.to_json_string(), feature_extractor.to_json_string())
+        self.assertEqual(processor_fast.feature_extractor.to_json_string(), feature_extractor.to_json_string())
+        self.assertIsInstance(processor_slow.feature_extractor, FlavaFeatureExtractor)
+        self.assertIsInstance(processor_fast.feature_extractor, FlavaFeatureExtractor)
+
+    def test_save_load_pretrained_additional_features(self):
+        processor = FlavaProcessor(tokenizer=self.get_tokenizer(), feature_extractor=self.get_feature_extractor())
+        processor.save_pretrained(self.tmpdirname)
+
+        tokenizer_add_kwargs = self.get_tokenizer(bos_token="(BOS)", eos_token="(EOS)")
+        feature_extractor_add_kwargs = self.get_feature_extractor(do_normalize=False, padding_value=1.0)
+
+        processor = FlavaProcessor.from_pretrained(
+            self.tmpdirname, bos_token="(BOS)", eos_token="(EOS)", do_normalize=False, padding_value=1.0
+        )
+
+        self.assertEqual(processor.tokenizer.get_vocab(), tokenizer_add_kwargs.get_vocab())
+        self.assertIsInstance(processor.tokenizer, BertTokenizerFast)
+
+        self.assertEqual(processor.feature_extractor.to_json_string(), feature_extractor_add_kwargs.to_json_string())
+        self.assertIsInstance(processor.feature_extractor, FlavaFeatureExtractor)
+
+    def test_feature_extractor(self):
+        feature_extractor = self.get_feature_extractor()
+        tokenizer = self.get_tokenizer()
+
+        processor = FlavaProcessor(tokenizer=tokenizer, feature_extractor=feature_extractor)
+
+        image_input = self.prepare_image_inputs()
+
+        input_feat_extract = feature_extractor(image_input, return_tensors="np")
+        input_processor = processor(images=image_input, return_tensors="np")
+
+        for key in input_feat_extract.keys():
+            self.assertAlmostEqual(input_feat_extract[key].sum(), input_processor[key].sum(), delta=1e-2)
+
+        # With rest of the args
+        random.seed(1234)
+        input_feat_extract = feature_extractor(
+            image_input, return_image_mask=True, return_codebook_pixels=True, return_tensors="np"
+        )
+        random.seed(1234)
+        input_processor = processor(
+            images=image_input, return_image_mask=True, return_codebook_pixels=True, return_tensors="np"
+        )
+
+        for key in input_feat_extract.keys():
+            self.assertAlmostEqual(input_feat_extract[key].sum(), input_processor[key].sum(), delta=1e-2)
+
+    def test_tokenizer(self):
+        feature_extractor = self.get_feature_extractor()
+        tokenizer = self.get_tokenizer()
+
+        processor = FlavaProcessor(tokenizer=tokenizer, feature_extractor=feature_extractor)
+
+        input_str = "lower newer"
+
+        encoded_processor = processor(text=input_str)
+
+        encoded_tok = tokenizer(input_str)
+
+        for key in encoded_tok.keys():
+            self.assertListEqual(encoded_tok[key], encoded_processor[key])
+
+    def test_processor(self):
+        feature_extractor = self.get_feature_extractor()
+        tokenizer = self.get_tokenizer()
+
+        processor = FlavaProcessor(tokenizer=tokenizer, feature_extractor=feature_extractor)
+
+        input_str = "lower newer"
+        image_input = self.prepare_image_inputs()
+
+        inputs = processor(text=input_str, images=image_input)
+
+        self.assertListEqual(list(inputs.keys()), ["input_ids", "token_type_ids", "attention_mask", "pixel_values"])
+
+        # add extra args
+        inputs = processor(text=input_str, images=image_input, return_codebook_pixels=True, return_image_mask=True)
+
+        self.assertListEqual(
+            list(inputs.keys()),
+            [
+                "input_ids",
+                "token_type_ids",
+                "attention_mask",
+                "pixel_values",
+                "codebook_pixel_values",
+                "bool_masked_pos",
+            ],
+        )
+
+        # test if it raises when no input is passed
+        with pytest.raises(ValueError):
+            processor()
+
+    def test_tokenizer_decode(self):
+        feature_extractor = self.get_feature_extractor()
+        tokenizer = self.get_tokenizer()
+
+        processor = FlavaProcessor(tokenizer=tokenizer, feature_extractor=feature_extractor)
+
+        predicted_ids = [[1, 4, 5, 8, 1, 0, 8], [3, 4, 3, 1, 1, 8, 9]]
+
+        decoded_processor = processor.batch_decode(predicted_ids)
+        decoded_tok = tokenizer.batch_decode(predicted_ids)
+
+        self.assertListEqual(decoded_tok, decoded_processor)

utils/check_repo.py

@@ -146,6 +146,10 @@ IGNORE_NON_AUTO_CONFIGURED = PRIVATE_MODELS.copy() + [
     "DetrForSegmentation",
     "DPRReader",
     "FlaubertForQuestionAnswering",
+    "FlavaImageCodebook",
+    "FlavaTextModel",
+    "FlavaImageModel",
+    "FlavaMultimodalModel",
     "GPT2DoubleHeadsModel",
     "LukeForMaskedLM",
     "LukeForEntityClassification",