Automatically sort auto mappings (#17250)

* Automatically sort auto mappings

* Better class extraction

* Some auto class magic

* Adapt test and underlying behavior

* Remove re-used config

* Quality

Author: sgugger

.circleci/config.yml

@@ -857,6 +857,7 @@ jobs:
             - run: black --check --preview examples tests src utils
             - run: isort --check-only examples tests src utils
             - run: python utils/custom_init_isort.py --check_only
+            - run: python utils/sort_auto_mappings.py --check_only
             - run: flake8 examples tests src utils
             - run: doc-builder style src/transformers docs/source --max_len 119 --check_only --path_to_docs docs/source
 

Makefile

@@ -48,13 +48,15 @@ quality:
 	black --check --preview $(check_dirs)
 	isort --check-only $(check_dirs)
 	python utils/custom_init_isort.py --check_only
+	python utils/sort_auto_mappings.py --check_only
 	flake8 $(check_dirs)
 	doc-builder style src/transformers docs/source --max_len 119 --check_only --path_to_docs docs/source
 
 # Format source code automatically and check is there are any problems left that need manual fixing
 
 extra_style_checks:
 	python utils/custom_init_isort.py
+	python utils/sort_auto_mappings.py
 	doc-builder style src/transformers docs/source --max_len 119 --path_to_docs docs/source
 
 # this target runs checks on all files and potentially modifies some of them

docs/source/en/index.mdx

@@ -259,7 +259,6 @@ Flax), PyTorch, and/or TensorFlow.
 |            Swin             |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
 |             T5              |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
 |            TAPAS            |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
-|            TAPEX            |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
 |       Transformer-XL        |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |
 |            TrOCR            |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
 |          UniSpeech          |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |

docs/source/en/serialization.mdx

@@ -74,7 +74,6 @@ Ready-made configurations include the following architectures:
 - RoBERTa
 - RoFormer
 - T5
-- TAPEX
 - ViT
 - XLM-RoBERTa
 - XLM-RoBERTa-XL

src/transformers/models/auto/auto_factory.py

@@ -560,10 +560,17 @@ def __getitem__(self, key):
         if key in self._extra_content:
             return self._extra_content[key]
         model_type = self._reverse_config_mapping[key.__name__]
-        if model_type not in self._model_mapping:
-            raise KeyError(key)
-        model_name = self._model_mapping[model_type]
-        return self._load_attr_from_module(model_type, model_name)
+        if model_type in self._model_mapping:
+            model_name = self._model_mapping[model_type]
+            return self._load_attr_from_module(model_type, model_name)
+
+        # Maybe there was several model types associated with this config.
+        model_types = [k for k, v in self._config_mapping.items() if v == key.__name__]
+        for mtype in model_types:
+            if mtype in self._model_mapping:
+                model_name = self._model_mapping[mtype]
+                return self._load_attr_from_module(mtype, model_name)
+        raise KeyError(key)
 
     def _load_attr_from_module(self, model_type, attr):
         module_name = model_type_to_module_name(model_type)

src/transformers/models/auto/configuration_auto.py

@@ -38,30 +38,30 @@
 FEATURE_EXTRACTOR_MAPPING_NAMES = OrderedDict(
     [
         ("beit", "BeitFeatureExtractor"),
-        ("detr", "DetrFeatureExtractor"),
-        ("deit", "DeiTFeatureExtractor"),
-        ("hubert", "Wav2Vec2FeatureExtractor"),
-        ("speech_to_text", "Speech2TextFeatureExtractor"),
-        ("vit", "ViTFeatureExtractor"),
-        ("wav2vec2", "Wav2Vec2FeatureExtractor"),
-        ("detr", "DetrFeatureExtractor"),
-        ("layoutlmv2", "LayoutLMv2FeatureExtractor"),
         ("clip", "CLIPFeatureExtractor"),
-        ("flava", "FlavaFeatureExtractor"),
-        ("perceiver", "PerceiverFeatureExtractor"),
-        ("swin", "ViTFeatureExtractor"),
-        ("vit_mae", "ViTFeatureExtractor"),
-        ("segformer", "SegformerFeatureExtractor"),
         ("convnext", "ConvNextFeatureExtractor"),
-        ("van", "ConvNextFeatureExtractor"),
-        ("resnet", "ConvNextFeatureExtractor"),
-        ("regnet", "ConvNextFeatureExtractor"),
-        ("poolformer", "PoolFormerFeatureExtractor"),
-        ("maskformer", "MaskFormerFeatureExtractor"),
         ("data2vec-audio", "Wav2Vec2FeatureExtractor"),
         ("data2vec-vision", "BeitFeatureExtractor"),
+        ("deit", "DeiTFeatureExtractor"),
+        ("detr", "DetrFeatureExtractor"),
+        ("detr", "DetrFeatureExtractor"),
         ("dpt", "DPTFeatureExtractor"),
+        ("flava", "FlavaFeatureExtractor"),
         ("glpn", "GLPNFeatureExtractor"),
+        ("hubert", "Wav2Vec2FeatureExtractor"),
+        ("layoutlmv2", "LayoutLMv2FeatureExtractor"),
+        ("maskformer", "MaskFormerFeatureExtractor"),
+        ("perceiver", "PerceiverFeatureExtractor"),
+        ("poolformer", "PoolFormerFeatureExtractor"),
+        ("regnet", "ConvNextFeatureExtractor"),
+        ("resnet", "ConvNextFeatureExtractor"),
+        ("segformer", "SegformerFeatureExtractor"),
+        ("speech_to_text", "Speech2TextFeatureExtractor"),
+        ("swin", "ViTFeatureExtractor"),
+        ("van", "ConvNextFeatureExtractor"),
+        ("vit", "ViTFeatureExtractor"),
+        ("vit_mae", "ViTFeatureExtractor"),
+        ("wav2vec2", "Wav2Vec2FeatureExtractor"),
         ("yolos", "YolosFeatureExtractor"),
     ]
 )
@@ -75,8 +75,10 @@ def feature_extractor_class_from_name(class_name: str):
             module_name = model_type_to_module_name(module_name)
 
             module = importlib.import_module(f".{module_name}", "transformers.models")
-            return getattr(module, class_name)
-            break
+            try:
+                return getattr(module, class_name)
+            except AttributeError:
+                continue
 
     for config, extractor in FEATURE_EXTRACTOR_MAPPING._extra_content.items():
         if getattr(extractor, "__name__", None) == class_name: