Fix init import_structure sorting (#20477)

* Fix init import_structure sorting

* Fix rebase

src/transformers/__init__.py

@@ -569,10 +569,10 @@ else:
     _import_structure["models.m2m_100"].append("M2M100Tokenizer")
     _import_structure["models.marian"].append("MarianTokenizer")
     _import_structure["models.mbart"].append("MBartTokenizer")
-    _import_structure["models.nllb"].append("NllbTokenizer")
     _import_structure["models.mbart50"].append("MBart50Tokenizer")
     _import_structure["models.mluke"].append("MLukeTokenizer")
     _import_structure["models.mt5"].append("MT5Tokenizer")
+    _import_structure["models.nllb"].append("NllbTokenizer")
     _import_structure["models.pegasus"].append("PegasusTokenizer")
     _import_structure["models.plbart"].append("PLBartTokenizer")
     _import_structure["models.reformer"].append("ReformerTokenizer")
@@ -722,14 +722,14 @@ else:
     _import_structure["image_utils"] = ["ImageFeatureExtractionMixin"]
     _import_structure["models.beit"].extend(["BeitFeatureExtractor", "BeitImageProcessor"])
     _import_structure["models.clip"].extend(["CLIPFeatureExtractor", "CLIPImageProcessor"])
+    _import_structure["models.conditional_detr"].append("ConditionalDetrFeatureExtractor")
     _import_structure["models.convnext"].extend(["ConvNextFeatureExtractor", "ConvNextImageProcessor"])
     _import_structure["models.deformable_detr"].append("DeformableDetrFeatureExtractor")
     _import_structure["models.deit"].extend(["DeiTFeatureExtractor", "DeiTImageProcessor"])
     _import_structure["models.detr"].append("DetrFeatureExtractor")
-    _import_structure["models.conditional_detr"].append("ConditionalDetrFeatureExtractor")
     _import_structure["models.donut"].extend(["DonutFeatureExtractor", "DonutImageProcessor"])
     _import_structure["models.dpt"].extend(["DPTFeatureExtractor", "DPTImageProcessor"])
-    _import_structure["models.flava"].extend(["FlavaFeatureExtractor", "FlavaProcessor", "FlavaImageProcessor"])
+    _import_structure["models.flava"].extend(["FlavaFeatureExtractor", "FlavaImageProcessor", "FlavaProcessor"])
     _import_structure["models.glpn"].extend(["GLPNFeatureExtractor", "GLPNImageProcessor"])
     _import_structure["models.imagegpt"].extend(["ImageGPTFeatureExtractor", "ImageGPTImageProcessor"])
     _import_structure["models.layoutlmv2"].extend(["LayoutLMv2FeatureExtractor", "LayoutLMv2ImageProcessor"])
@@ -819,70 +819,44 @@ else:
         "TextDatasetForNextSentencePrediction",
     ]
     _import_structure["deepspeed"] = []
-    _import_structure["generation_utils"] = []
     _import_structure["generation"].extend(
         [
-            "Constraint",
-            "ConstraintListState",
-            "DisjunctiveConstraint",
-            "PhrasalConstraint",
             "BeamScorer",
             "BeamSearchScorer",
             "ConstrainedBeamSearchScorer",
+            "Constraint",
+            "ConstraintListState",
+            "DisjunctiveConstraint",
             "ForcedBOSTokenLogitsProcessor",
             "ForcedEOSTokenLogitsProcessor",
+            "GenerationMixin",
             "HammingDiversityLogitsProcessor",
             "InfNanRemoveLogitsProcessor",
             "LogitsProcessor",
             "LogitsProcessorList",
             "LogitsWarper",
+            "MaxLengthCriteria",
+            "MaxTimeCriteria",
             "MinLengthLogitsProcessor",
             "NoBadWordsLogitsProcessor",
             "NoRepeatNGramLogitsProcessor",
+            "PhrasalConstraint",
             "PrefixConstrainedLogitsProcessor",
             "RepetitionPenaltyLogitsProcessor",
+            "StoppingCriteria",
+            "StoppingCriteriaList",
             "TemperatureLogitsWarper",
             "TopKLogitsWarper",
             "TopPLogitsWarper",
             "TypicalLogitsWarper",
-            "MaxLengthCriteria",
-            "MaxTimeCriteria",
-            "StoppingCriteria",
-            "StoppingCriteriaList",
-            "GenerationMixin",
             "top_k_top_p_filtering",
         ]
     )
+    _import_structure["generation_utils"] = []
     _import_structure["modeling_outputs"] = []
     _import_structure["modeling_utils"] = ["PreTrainedModel"]
 
     # PyTorch models structure
-
-    _import_structure["models.roc_bert"].extend(
-        [
-            "ROC_BERT_PRETRAINED_MODEL_ARCHIVE_LIST",
-            "RoCBertForMaskedLM",
-            "RoCBertForCausalLM",
-            "RoCBertForMultipleChoice",
-            "RoCBertForQuestionAnswering",
-            "RoCBertForSequenceClassification",
-            "RoCBertForTokenClassification",
-            "RoCBertLayer",
-            "RoCBertModel",
-            "RoCBertForPreTraining",
-            "RoCBertPreTrainedModel",
-            "load_tf_weights_in_roc_bert",
-        ]
-    )
-
-    _import_structure["models.time_series_transformer"].extend(
-        [
-            "TIME_SERIES_TRANSFORMER_PRETRAINED_MODEL_ARCHIVE_LIST",
-            "TimeSeriesTransformerForPrediction",
-            "TimeSeriesTransformerModel",
-            "TimeSeriesTransformerPreTrainedModel",
-        ]
-    )
     _import_structure["models.albert"].extend(
         [
             "ALBERT_PRETRAINED_MODEL_ARCHIVE_LIST",
@@ -897,12 +871,13 @@ else:
             "load_tf_weights_in_albert",
         ]
     )
+
     _import_structure["models.audio_spectrogram_transformer"].extend(
         [
             "AUDIO_SPECTROGRAM_TRANSFORMER_PRETRAINED_MODEL_ARCHIVE_LIST",
+            "ASTForAudioClassification",
             "ASTModel",
             "ASTPreTrainedModel",
-            "ASTForAudioClassification",
         ]
     )
     _import_structure["models.auto"].extend(
@@ -913,8 +888,8 @@ else:
             "MODEL_FOR_CAUSAL_IMAGE_MODELING_MAPPING",
             "MODEL_FOR_CAUSAL_LM_MAPPING",
             "MODEL_FOR_CTC_MAPPING",
-            "MODEL_FOR_DOCUMENT_QUESTION_ANSWERING_MAPPING",
             "MODEL_FOR_DEPTH_ESTIMATION_MAPPING",
+            "MODEL_FOR_DOCUMENT_QUESTION_ANSWERING_MAPPING",
             "MODEL_FOR_IMAGE_CLASSIFICATION_MAPPING",
             "MODEL_FOR_IMAGE_SEGMENTATION_MAPPING",
             "MODEL_FOR_INSTANCE_SEGMENTATION_MAPPING",
@@ -934,18 +909,18 @@ else:
             "MODEL_FOR_VIDEO_CLASSIFICATION_MAPPING",
             "MODEL_FOR_VISION_2_SEQ_MAPPING",
             "MODEL_FOR_VISUAL_QUESTION_ANSWERING_MAPPING",
+            "MODEL_FOR_ZERO_SHOT_OBJECT_DETECTION_MAPPING",
             "MODEL_MAPPING",
             "MODEL_WITH_LM_HEAD_MAPPING",
-            "MODEL_FOR_ZERO_SHOT_OBJECT_DETECTION_MAPPING",
-            "AutoModel",
             "AutoBackbone",
+            "AutoModel",
             "AutoModelForAudioClassification",
             "AutoModelForAudioFrameClassification",
             "AutoModelForAudioXVector",
             "AutoModelForCausalLM",
             "AutoModelForCTC",
-            "AutoModelForDocumentQuestionAnswering",
             "AutoModelForDepthEstimation",
+            "AutoModelForDocumentQuestionAnswering",
             "AutoModelForImageClassification",
             "AutoModelForImageSegmentation",
             "AutoModelForInstanceSegmentation",
@@ -965,8 +940,8 @@ else:
             "AutoModelForVideoClassification",
             "AutoModelForVision2Seq",
             "AutoModelForVisualQuestionAnswering",
-            "AutoModelWithLMHead",
             "AutoModelForZeroShotObjectDetection",
+            "AutoModelWithLMHead",
         ]
     )
     _import_structure["models.bart"].extend(
@@ -981,17 +956,6 @@ else:
             "PretrainedBartModel",
         ]
     )
-    _import_structure["models.mvp"].extend(
-        [
-            "MVP_PRETRAINED_MODEL_ARCHIVE_LIST",
-            "MvpForCausalLM",
-            "MvpForConditionalGeneration",
-            "MvpForQuestionAnswering",
-            "MvpForSequenceClassification",
-            "MvpModel",
-            "MvpPreTrainedModel",
-        ]
-    )
     _import_structure["models.beit"].extend(
         [
             "BEIT_PRETRAINED_MODEL_ARCHIVE_LIST",
@@ -1054,17 +1018,6 @@ else:
             "BigBirdPegasusPreTrainedModel",
         ]
     )
-    _import_structure["models.bloom"].extend(
-        [
-            "BLOOM_PRETRAINED_MODEL_ARCHIVE_LIST",
-            "BloomForCausalLM",
-            "BloomModel",
-            "BloomPreTrainedModel",
-            "BloomForSequenceClassification",
-            "BloomForTokenClassification",
-            "BloomForQuestionAnswering",
-        ]
-    )
     _import_structure["models.blenderbot"].extend(
         [
             "BLENDERBOT_PRETRAINED_MODEL_ARCHIVE_LIST",
@@ -1083,6 +1036,17 @@ else:
             "BlenderbotSmallPreTrainedModel",
         ]
     )
+    _import_structure["models.bloom"].extend(
+        [
+            "BLOOM_PRETRAINED_MODEL_ARCHIVE_LIST",
+            "BloomForCausalLM",
+            "BloomForQuestionAnswering",
+            "BloomForSequenceClassification",
+            "BloomForTokenClassification",
+            "BloomModel",
+            "BloomPreTrainedModel",
+        ]
+    )
     _import_structure["models.camembert"].extend(
         [
             "CAMEMBERT_PRETRAINED_MODEL_ARCHIVE_LIST",
@@ -1123,20 +1087,19 @@ else:
     _import_structure["models.clipseg"].extend(
         [
             "CLIPSEG_PRETRAINED_MODEL_ARCHIVE_LIST",
+            "CLIPSegForImageSegmentation",
             "CLIPSegModel",
             "CLIPSegPreTrainedModel",
             "CLIPSegTextModel",
             "CLIPSegVisionModel",
-            "CLIPSegForImageSegmentation",
         ]
     )
-    _import_structure["models.x_clip"].extend(
+    _import_structure["models.codegen"].extend(
         [
-            "XCLIP_PRETRAINED_MODEL_ARCHIVE_LIST",
-            "XCLIPModel",
-            "XCLIPPreTrainedModel",
-            "XCLIPTextModel",
-            "XCLIPVisionModel",
+            "CODEGEN_PRETRAINED_MODEL_ARCHIVE_LIST",
+            "CodeGenForCausalLM",
+            "CodeGenModel",
+            "CodeGenPreTrainedModel",
         ]
     )
     _import_structure["models.convbert"].extend(
@@ -1245,6 +1208,14 @@ else:
             "DeiTPreTrainedModel",
         ]
     )
+    _import_structure["models.dinat"].extend(
+        [
+            "DINAT_PRETRAINED_MODEL_ARCHIVE_LIST",
+            "DinatForImageClassification",
+            "DinatModel",
+            "DinatPreTrainedModel",
+        ]
+    )
     _import_structure["models.distilbert"].extend(
         [
             "DISTILBERT_PRETRAINED_MODEL_ARCHIVE_LIST",
@@ -1257,14 +1228,6 @@ else:
             "DistilBertPreTrainedModel",
         ]
     )
-    _import_structure["models.dinat"].extend(
-        [
-            "DINAT_PRETRAINED_MODEL_ARCHIVE_LIST",
-            "DinatForImageClassification",
-            "DinatModel",
-            "DinatPreTrainedModel",
-        ]
-    )
     _import_structure["models.donut"].extend(
         [
             "DONUT_SWIN_PRETRAINED_MODEL_ARCHIVE_LIST",
@@ -1347,8 +1310,8 @@ else:
             "FlaubertForSequenceClassification",
             "FlaubertForTokenClassification",
             "FlaubertModel",
-            "FlaubertWithLMHeadModel",
             "FlaubertPreTrainedModel",
+            "FlaubertWithLMHeadModel",
         ]
     )
     _import_structure["models.flava"].extend(
@@ -1461,14 +1424,6 @@ else:
             "GroupViTVisionModel",
         ]
     )
-    _import_structure["models.codegen"].extend(
-        [
-            "CODEGEN_PRETRAINED_MODEL_ARCHIVE_LIST",
-            "CodeGenForCausalLM",
-            "CodeGenModel",
-            "CodeGenPreTrainedModel",
-        ]
-    )
     _import_structure["models.hubert"].extend(
         [
             "HUBERT_PRETRAINED_MODEL_ARCHIVE_LIST",
@@ -1505,17 +1460,17 @@ else:
             "JUKEBOX_PRETRAINED_MODEL_ARCHIVE_LIST",
             "JukeboxModel",
             "JukeboxPreTrainedModel",
-            "JukeboxVQVAE",
             "JukeboxPrior",
+            "JukeboxVQVAE",
         ]
     )
     _import_structure["models.layoutlm"].extend(
         [
             "LAYOUTLM_PRETRAINED_MODEL_ARCHIVE_LIST",
             "LayoutLMForMaskedLM",
+            "LayoutLMForQuestionAnswering",
             "LayoutLMForSequenceClassification",
             "LayoutLMForTokenClassification",
-            "LayoutLMForQuestionAnswering",
             "LayoutLMModel",
             "LayoutLMPreTrainedModel",
         ]
@@ -1559,6 +1514,16 @@ else:
             "LevitPreTrainedModel",
         ]
     )
+    _import_structure["models.lilt"].extend(
+        [
+            "LILT_PRETRAINED_MODEL_ARCHIVE_LIST",
+            "LiltForQuestionAnswering",
+            "LiltForSequenceClassification",
+            "LiltForTokenClassification",
+            "LiltModel",
+            "LiltPreTrainedModel",
+        ]
+    )
     _import_structure["models.longformer"].extend(
         [
             "LONGFORMER_PRETRAINED_MODEL_ARCHIVE_LIST",
@@ -1587,11 +1552,11 @@ else:
             "LukeForEntityClassification",
             "LukeForEntityPairClassification",
             "LukeForEntitySpanClassification",
+            "LukeForMaskedLM",
             "LukeForMultipleChoice",
             "LukeForQuestionAnswering",
             "LukeForSequenceClassification",
             "LukeForTokenClassification",
-            "LukeForMaskedLM",
             "LukeModel",
             "LukePreTrainedModel",
         ]
@@ -1616,15 +1581,6 @@ else:
         ]
     )
     _import_structure["models.marian"].extend(["MarianForCausalLM", "MarianModel", "MarianMTModel"])
-    _import_structure["models.maskformer"].extend(
-        [
-            "MASKFORMER_PRETRAINED_MODEL_ARCHIVE_LIST",
-            "MaskFormerForInstanceSegmentation",
-            "MaskFormerModel",
-            "MaskFormerPreTrainedModel",
-            "MaskFormerSwinBackbone",
-        ]
-    )
     _import_structure["models.markuplm"].extend(
         [
             "MARKUPLM_PRETRAINED_MODEL_ARCHIVE_LIST",
@@ -1635,6 +1591,15 @@ else:
             "MarkupLMPreTrainedModel",
         ]
     )
+    _import_structure["models.maskformer"].extend(
+        [
+            "MASKFORMER_PRETRAINED_MODEL_ARCHIVE_LIST",
+            "MaskFormerForInstanceSegmentation",
+            "MaskFormerModel",
+            "MaskFormerPreTrainedModel",
+            "MaskFormerSwinBackbone",
+        ]
+    )
     _import_structure["models.mbart"].extend(
         [
             "MBartForCausalLM",
@@ -1727,6 +1692,17 @@ else:
         ]
     )
     _import_structure["models.mt5"].extend(["MT5EncoderModel", "MT5ForConditionalGeneration", "MT5Model"])
+    _import_structure["models.mvp"].extend(
+        [
+            "MVP_PRETRAINED_MODEL_ARCHIVE_LIST",
+            "MvpForCausalLM",
+            "MvpForConditionalGeneration",
+            "MvpForQuestionAnswering",
+            "MvpForSequenceClassification",
+            "MvpModel",
+            "MvpPreTrainedModel",
+        ]
+    )
     _import_structure["models.nat"].extend(
         [
             "NAT_PRETRAINED_MODEL_ARCHIVE_LIST",
@@ -1739,9 +1715,9 @@ else:
         [
             "NEZHA_PRETRAINED_MODEL_ARCHIVE_LIST",
             "NezhaForMaskedLM",
-            "NezhaForPreTraining",
-            "NezhaForNextSentencePrediction",
             "NezhaForMultipleChoice",
+            "NezhaForNextSentencePrediction",
+            "NezhaForPreTraining",
             "NezhaForQuestionAnswering",
             "NezhaForSequenceClassification",
             "NezhaForTokenClassification",
@@ -1777,20 +1753,20 @@ else:
         [
             "OPT_PRETRAINED_MODEL_ARCHIVE_LIST",
             "OPTForCausalLM",
+            "OPTForQuestionAnswering",
+            "OPTForSequenceClassification",
             "OPTModel",
             "OPTPreTrainedModel",
-            "OPTForSequenceClassification",
-            "OPTForQuestionAnswering",
         ]
     )
     _import_structure["models.owlvit"].extend(
         [
             "OWLVIT_PRETRAINED_MODEL_ARCHIVE_LIST",
+            "OwlViTForObjectDetection",
             "OwlViTModel",
             "OwlViTPreTrainedModel",
             "OwlViTTextModel",
             "OwlViTVisionModel",
-            "OwlViTForObjectDetection",
         ]
     )
     _import_structure["models.pegasus"].extend(
@@ -1919,10 +1895,10 @@ else:
     _import_structure["models.resnet"].extend(
         [
             "RESNET_PRETRAINED_MODEL_ARCHIVE_LIST",
+            "ResNetBackbone",
             "ResNetForImageClassification",
             "ResNetModel",
             "ResNetPreTrainedModel",
-            "ResNetBackbone",
         ]
     )
     _import_structure["models.retribert"].extend(
@@ -1941,14 +1917,20 @@ else:
             "RobertaPreTrainedModel",
         ]
     )
-    _import_structure["models.lilt"].extend(
+    _import_structure["models.roc_bert"].extend(
         [
-            "LILT_PRETRAINED_MODEL_ARCHIVE_LIST",
-            "LiltForQuestionAnswering",
-            "LiltForSequenceClassification",
-            "LiltForTokenClassification",
-            "LiltModel",
-            "LiltPreTrainedModel",
+            "ROC_BERT_PRETRAINED_MODEL_ARCHIVE_LIST",
+            "RoCBertForCausalLM",
+            "RoCBertForMaskedLM",
+            "RoCBertForMultipleChoice",
+            "RoCBertForPreTraining",
+            "RoCBertForQuestionAnswering",
+            "RoCBertForSequenceClassification",
+            "RoCBertForTokenClassification",
+            "RoCBertLayer",
+            "RoCBertModel",
+            "RoCBertPreTrainedModel",
+            "load_tf_weights_in_roc_bert",
         ]
     )
     _import_structure["models.roformer"].extend(
@@ -2004,14 +1986,6 @@ else:
             "Speech2TextPreTrainedModel",
         ]
     )
-    _import_structure["models.whisper"].extend(
-        [
-            "WHISPER_PRETRAINED_MODEL_ARCHIVE_LIST",
-            "WhisperForConditionalGeneration",
-            "WhisperModel",
-            "WhisperPreTrainedModel",
-        ]
-    )
     _import_structure["models.speech_to_text_2"].extend(["Speech2Text2ForCausalLM", "Speech2Text2PreTrainedModel"])
     _import_structure["models.splinter"].extend(
         [
@@ -2054,15 +2028,15 @@ else:
             "Swinv2PreTrainedModel",
         ]
     )
-    _import_structure["models.tapas"].extend(
+    _import_structure["models.switch_transformers"].extend(
         [
-            "TAPAS_PRETRAINED_MODEL_ARCHIVE_LIST",
-            "TapasForMaskedLM",
-            "TapasForQuestionAnswering",
-            "TapasForSequenceClassification",
-            "TapasModel",
-            "TapasPreTrainedModel",
-            "load_tf_weights_in_tapas",
+            "SWITCH_TRANSFORMERS_PRETRAINED_MODEL_ARCHIVE_LIST",
+            "SwitchTransformersEncoderModel",
+            "SwitchTransformersForConditionalGeneration",
+            "SwitchTransformersModel",
+            "SwitchTransformersPreTrainedModel",
+            "SwitchTransformersSparseMLP",
+            "SwitchTransformersTop1Router",
         ]
     )
     _import_structure["models.t5"].extend(
@@ -2075,15 +2049,23 @@ else:
             "load_tf_weights_in_t5",
         ]
     )
-    _import_structure["models.switch_transformers"].extend(
+    _import_structure["models.tapas"].extend(
         [
-            "SWITCH_TRANSFORMERS_PRETRAINED_MODEL_ARCHIVE_LIST",
-            "SwitchTransformersEncoderModel",
-            "SwitchTransformersForConditionalGeneration",
-            "SwitchTransformersModel",
-            "SwitchTransformersPreTrainedModel",
-            "SwitchTransformersTop1Router",
-            "SwitchTransformersSparseMLP",
+            "TAPAS_PRETRAINED_MODEL_ARCHIVE_LIST",
+            "TapasForMaskedLM",
+            "TapasForQuestionAnswering",
+            "TapasForSequenceClassification",
+            "TapasModel",
+            "TapasPreTrainedModel",
+            "load_tf_weights_in_tapas",
+        ]
+    )
+    _import_structure["models.time_series_transformer"].extend(
+        [
+            "TIME_SERIES_TRANSFORMER_PRETRAINED_MODEL_ARCHIVE_LIST",
+            "TimeSeriesTransformerForPrediction",
+            "TimeSeriesTransformerModel",
+            "TimeSeriesTransformerPreTrainedModel",
         ]
     )
     _import_structure["models.trajectory_transformer"].extend(
@@ -2137,14 +2119,23 @@ else:
             "VanPreTrainedModel",
         ]
     )
+    _import_structure["models.videomae"].extend(
+        [
+            "VIDEOMAE_PRETRAINED_MODEL_ARCHIVE_LIST",
+            "VideoMAEForPreTraining",
+            "VideoMAEForVideoClassification",
+            "VideoMAEModel",
+            "VideoMAEPreTrainedModel",
+        ]
+    )
     _import_structure["models.vilt"].extend(
         [
             "VILT_PRETRAINED_MODEL_ARCHIVE_LIST",
             "ViltForImageAndTextRetrieval",
             "ViltForImagesAndTextClassification",
-            "ViltForTokenClassification",
             "ViltForMaskedLM",
             "ViltForQuestionAnswering",
+            "ViltForTokenClassification",
             "ViltLayer",
             "ViltModel",
             "ViltPreTrainedModel",
@@ -2186,20 +2177,11 @@ else:
     _import_structure["models.vit_msn"].extend(
         [
             "VIT_MSN_PRETRAINED_MODEL_ARCHIVE_LIST",
-            "ViTMSNModel",
             "ViTMSNForImageClassification",
+            "ViTMSNModel",
             "ViTMSNPreTrainedModel",
         ]
     )
-    _import_structure["models.videomae"].extend(
-        [
-            "VIDEOMAE_PRETRAINED_MODEL_ARCHIVE_LIST",
-            "VideoMAEForPreTraining",
-            "VideoMAEModel",
-            "VideoMAEPreTrainedModel",
-            "VideoMAEForVideoClassification",
-        ]
-    )
     _import_structure["models.wav2vec2"].extend(
         [
             "WAV_2_VEC_2_PRETRAINED_MODEL_ARCHIVE_LIST",
@@ -2236,6 +2218,23 @@ else:
             "WavLMPreTrainedModel",
         ]
     )
+    _import_structure["models.whisper"].extend(
+        [
+            "WHISPER_PRETRAINED_MODEL_ARCHIVE_LIST",
+            "WhisperForConditionalGeneration",
+            "WhisperModel",
+            "WhisperPreTrainedModel",
+        ]
+    )
+    _import_structure["models.x_clip"].extend(
+        [
+            "XCLIP_PRETRAINED_MODEL_ARCHIVE_LIST",
+            "XCLIPModel",
+            "XCLIPPreTrainedModel",
+            "XCLIPTextModel",
+            "XCLIPVisionModel",
+        ]
+    )
     _import_structure["models.xglm"].extend(
         [
             "XGLM_PRETRAINED_MODEL_ARCHIVE_LIST",
@@ -2358,11 +2357,11 @@ else:
     _import_structure["activations_tf"] = []
     _import_structure["benchmark.benchmark_args_tf"] = ["TensorFlowBenchmarkArguments"]
     _import_structure["benchmark.benchmark_tf"] = ["TensorFlowBenchmark"]
-    _import_structure["generation_tf_utils"] = []
     _import_structure["generation"].extend(
         [
             "TFForcedBOSTokenLogitsProcessor",
             "TFForcedEOSTokenLogitsProcessor",
+            "TFGenerationMixin",
             "TFLogitsProcessor",
             "TFLogitsProcessorList",
             "TFLogitsWarper",
@@ -2373,10 +2372,10 @@ else:
             "TFTemperatureLogitsWarper",
             "TFTopKLogitsWarper",
             "TFTopPLogitsWarper",
-            "TFGenerationMixin",
             "tf_top_k_top_p_filtering",
         ]
     )
+    _import_structure["generation_tf_utils"] = []
     _import_structure["keras_callbacks"] = ["KerasMetricCallback", "PushToHubCallback"]
     _import_structure["modeling_tf_outputs"] = []
     _import_structure["modeling_tf_utils"] = [
@@ -2403,13 +2402,13 @@ else:
     _import_structure["models.auto"].extend(
         [
             "TF_MODEL_FOR_CAUSAL_LM_MAPPING",
+            "TF_MODEL_FOR_DOCUMENT_QUESTION_ANSWERING_MAPPING",
             "TF_MODEL_FOR_IMAGE_CLASSIFICATION_MAPPING",
             "TF_MODEL_FOR_MASKED_IMAGE_MODELING_MAPPING",
             "TF_MODEL_FOR_MASKED_LM_MAPPING",
             "TF_MODEL_FOR_MULTIPLE_CHOICE_MAPPING",
             "TF_MODEL_FOR_NEXT_SENTENCE_PREDICTION_MAPPING",
             "TF_MODEL_FOR_PRETRAINING_MAPPING",
-            "TF_MODEL_FOR_DOCUMENT_QUESTION_ANSWERING_MAPPING",
             "TF_MODEL_FOR_QUESTION_ANSWERING_MAPPING",
             "TF_MODEL_FOR_SEMANTIC_SEGMENTATION_MAPPING",
             "TF_MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING",
@@ -2422,12 +2421,12 @@ else:
             "TF_MODEL_WITH_LM_HEAD_MAPPING",
             "TFAutoModel",
             "TFAutoModelForCausalLM",
+            "TFAutoModelForDocumentQuestionAnswering",
             "TFAutoModelForImageClassification",
             "TFAutoModelForMaskedLM",
             "TFAutoModelForMultipleChoice",
             "TFAutoModelForNextSentencePrediction",
             "TFAutoModelForPreTraining",
-            "TFAutoModelForDocumentQuestionAnswering",
             "TFAutoModelForQuestionAnswering",
             "TFAutoModelForSemanticSegmentation",
             "TFAutoModelForSeq2SeqLM",
@@ -2679,8 +2678,8 @@ else:
         [
             "TF_LAYOUTLM_PRETRAINED_MODEL_ARCHIVE_LIST",
             "TFLayoutLMForMaskedLM",
-            "TFLayoutLMForSequenceClassification",
             "TFLayoutLMForQuestionAnswering",
+            "TFLayoutLMForSequenceClassification",
             "TFLayoutLMForTokenClassification",
             "TFLayoutLMMainLayer",
             "TFLayoutLMModel",
@@ -2743,10 +2742,10 @@ else:
     _import_structure["models.mobilevit"].extend(
         [
             "TF_MOBILEVIT_PRETRAINED_MODEL_ARCHIVE_LIST",
-            "TFMobileViTPreTrainedModel",
-            "TFMobileViTModel",
             "TFMobileViTForImageClassification",
             "TFMobileViTForSemanticSegmentation",
+            "TFMobileViTModel",
+            "TFMobileViTPreTrainedModel",
         ]
     )
     _import_structure["models.mpnet"].extend(
@@ -2999,11 +2998,11 @@ except OptionalDependencyNotAvailable:
         name for name in dir(dummy_flax_objects) if not name.startswith("_")
     ]
 else:
-    _import_structure["generation_flax_utils"] = []
     _import_structure["generation"].extend(
         [
             "FlaxForcedBOSTokenLogitsProcessor",
             "FlaxForcedEOSTokenLogitsProcessor",
+            "FlaxGenerationMixin",
             "FlaxLogitsProcessor",
             "FlaxLogitsProcessorList",
             "FlaxLogitsWarper",
@@ -3011,9 +3010,9 @@ else:
             "FlaxTemperatureLogitsWarper",
             "FlaxTopKLogitsWarper",
             "FlaxTopPLogitsWarper",
-            "FlaxGenerationMixin",
         ]
     )
+    _import_structure["generation_flax_utils"] = []
     _import_structure["modeling_flax_outputs"] = []
     _import_structure["modeling_flax_utils"] = ["FlaxPreTrainedModel"]
     _import_structure["models.albert"].extend(

src/transformers/models/speech_to_text/__init__.py

@@ -47,8 +47,13 @@ except OptionalDependencyNotAvailable:
 else:
     _import_structure["feature_extraction_speech_to_text"] = ["Speech2TextFeatureExtractor"]
 
-    if is_sentencepiece_available():
-        _import_structure["processing_speech_to_text"] = ["Speech2TextProcessor"]
+try:
+    if not (is_speech_available() and is_sentencepiece_available()):
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
+    _import_structure["processing_speech_to_text"] = ["Speech2TextProcessor"]
 
 try:
     if not is_tf_available():
@@ -96,8 +101,13 @@ if TYPE_CHECKING:
     else:
         from .feature_extraction_speech_to_text import Speech2TextFeatureExtractor
 
-        if is_sentencepiece_available():
-            from .processing_speech_to_text import Speech2TextProcessor
+    try:
+        if not (is_speech_available() and is_sentencepiece_available()):
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
+        from .processing_speech_to_text import Speech2TextProcessor
 
     try:
         if not is_tf_available():

utils/custom_init_isort.py

@@ -200,9 +200,9 @@ def sort_imports(file, check_only=True):
         indent = get_indent(block_lines[1])
         # Slit the internal block into blocks of indent level 1.
         internal_blocks = split_code_in_indented_blocks(internal_block_code, indent_level=indent)
-        # We have two categories of import key: list or _import_structu[key].append/extend
-        pattern = _re_direct_key if "_import_structure" in block_lines[0] else _re_indirect_key
-        # Grab the keys, but there is a trap: some lines are empty or jsut comments.
+        # We have two categories of import key: list or _import_structure[key].append/extend
+        pattern = _re_direct_key if "_import_structure = {" in block_lines[0] else _re_indirect_key
+        # Grab the keys, but there is a trap: some lines are empty or just comments.
         keys = [(pattern.search(b).groups()[0] if pattern.search(b) is not None else None) for b in internal_blocks]
         # We only sort the lines with a key.
         keys_to_sort = [(i, key) for i, key in enumerate(keys) if key is not None]