Add Data2Vec for Vision in TF

docs/source/en/index.mdx

@@ -191,7 +191,7 @@ Flax), PyTorch, and/or TensorFlow.
 |            CTRL             |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
 |        Data2VecAudio        |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
 |        Data2VecText         |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
-|       Data2VecVision        |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
+|       Data2VecVision        |       ❌       |       ❌       |       ✅        |         ✅         |      ❌      |
 |           DeBERTa           |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
 |         DeBERTa-v2          |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
 |    Decision Transformer     |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |

docs/source/en/model_doc/data2vec.mdx

@@ -38,9 +38,11 @@ Tips:
 - For Data2VecText, preprocessing is identical to [`RobertaModel`], including tokenization.
 - For Data2VecVision, preprocessing is identical to [`BeitModel`], including feature extraction.
 
-This model was contributed by [edugp](https://huggingface.co/edugp) and [patrickvonplaten](https://huggingface.co/patrickvonplaten)
+This model was contributed by [edugp](https://huggingface.co/edugp) and [patrickvonplaten](https://huggingface.co/patrickvonplaten).
+[sayakpaul](https://github.com/sayakpaul) contributed Data2Vec for vision in TensorFlow.
 
-The original code can be found [here](https://github.com/pytorch/fairseq/tree/main/examples/data2vec).
+The original code (for NLP and Speech) can be found [here](https://github.com/pytorch/fairseq/tree/main/examples/data2vec).
+The original code for vision can be found [here](https://github.com/facebookresearch/data2vec_vision/tree/main/beit).
 
 
 ## Data2VecTextConfig
@@ -130,3 +132,13 @@ The original code can be found [here](https://github.com/pytorch/fairseq/tree/ma
 
 [[autodoc]] Data2VecVisionForSemanticSegmentation
     - forward
+
+## TFData2VecVisionModel
+
+[[autodoc]] TFData2VecVisionModel
+    - call
+
+## TFData2VecVisionForImageClassification
+
+[[autodoc]] TFData2VecVisionForImageClassification
+    - call

src/transformers/__init__.py

@@ -1878,6 +1878,13 @@
             "TFCTRLPreTrainedModel",
         ]
     )
+    _import_structure["models.data2vec"].extend(
+        [
+            "TFData2VecVisionForImageClassification",
+            "TFData2VecVisionModel",
+            "TFData2VecVisionPreTrainedModel",
+        ]
+    )
     _import_structure["models.deberta"].extend(
         [
             "TF_DEBERTA_PRETRAINED_MODEL_ARCHIVE_LIST",
@@ -4029,6 +4036,11 @@
             TFCTRLModel,
             TFCTRLPreTrainedModel,
         )
+        from .models.data2vec import (
+            TFData2VecVisionForImageClassification,
+            TFData2VecVisionModel,
+            TFData2VecVisionPreTrainedModel,
+        )
         from .models.deberta import (
             TF_DEBERTA_PRETRAINED_MODEL_ARCHIVE_LIST,
             TFDebertaForMaskedLM,

src/transformers/models/auto/modeling_tf_auto.py

@@ -37,6 +37,7 @@
         ("roformer", "TFRoFormerModel"),
         ("convbert", "TFConvBertModel"),
         ("convnext", "TFConvNextModel"),
+        ("data2vec-vision", "TFData2VecVisionModel"),
         ("led", "TFLEDModel"),
         ("lxmert", "TFLxmertModel"),
         ("mt5", "TFMT5Model"),
@@ -163,6 +164,7 @@
         # Model for Image-classsification
         ("vit", "TFViTForImageClassification"),
         ("convnext", "TFConvNextForImageClassification"),
+        ("data2vec-vision", "TFData2VecVisionForImageClassification"),
     ]
 )
 

src/transformers/models/data2vec/__init__.py

@@ -18,6 +18,8 @@
 
 from typing import TYPE_CHECKING
 
+from transformers.utils.import_utils import is_tf_available
+
 from ...utils import _LazyModule, is_torch_available
 
 
@@ -68,6 +70,13 @@
         "Data2VecVisionPreTrainedModel",
     ]
 
+if is_tf_available():
+    _import_structure["modeling_tf_data2vec_vision"] = [
+        "TFData2VecVisionForImageClassification",
+        "TFData2VecVisionModel",
+        "TFData2VecVisionPreTrainedModel",
+    ]
+
 if TYPE_CHECKING:
     from .configuration_data2vec_audio import DATA2VEC_AUDIO_PRETRAINED_CONFIG_ARCHIVE_MAP, Data2VecAudioConfig
     from .configuration_data2vec_text import (
@@ -110,6 +119,12 @@
             Data2VecVisionModel,
             Data2VecVisionPreTrainedModel,
         )
+    if is_tf_available():
+        from .modeling_tf_data2vec_vision import (
+            TFData2VecVisionForImageClassification,
+            TFData2VecVisionModel,
+            TFData2VecVisionPreTrainedModel,
+        )
 
 else:
     import sys