[WIP] Fix Pyright static type checking by replacing if-else imports with try-except (#16578)

* rebase and isort

* modify cookiecutter init

* fix cookiecutter auto imports

* fix clean_frameworks_in_init

* fix add_model_to_main_init

* blackify

* replace unnecessary f-strings

* update yolos imports

* fix roberta import bug

* fix yolos missing dependency

* fix add_model_like and cookiecutter bug

* fix repository consistency error

* modify cookiecutter, fix add_new_model_like

* remove stale line
Co-authored-by: Dom Miketa <dmiketa@exscientia.co.uk>

src/transformers/__init__.py

@@ -29,6 +29,7 @@ from typing import TYPE_CHECKING
 # Check the dependencies satisfy the minimal versions required.
 from . import dependency_versions_check
 from .utils import (
+    OptionalDependencyNotAvailable,
     _LazyModule,
     is_flax_available,
     is_scatter_available,
@@ -412,7 +413,16 @@ _import_structure = {
 }
 
 # sentencepiece-backed objects
-if is_sentencepiece_available():
+try:
+    if not is_sentencepiece_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    from .utils import dummy_sentencepiece_objects
+
+    _import_structure["utils.dummy_sentencepiece_objects"] = [
+        name for name in dir(dummy_sentencepiece_objects) if not name.startswith("_")
+    ]
+else:
     _import_structure["models.albert"].append("AlbertTokenizer")
     _import_structure["models.barthez"].append("BarthezTokenizer")
     _import_structure["models.bartpho"].append("BartphoTokenizer")
@@ -439,16 +449,19 @@ if is_sentencepiece_available():
     _import_structure["models.xlm_prophetnet"].append("XLMProphetNetTokenizer")
     _import_structure["models.xlm_roberta"].append("XLMRobertaTokenizer")
     _import_structure["models.xlnet"].append("XLNetTokenizer")
-else:
-    from .utils import dummy_sentencepiece_objects
-
-    _import_structure["utils.dummy_sentencepiece_objects"] = [
-        name for name in dir(dummy_sentencepiece_objects) if not name.startswith("_")
-    ]
 
 # tokenizers-backed objects
-if is_tokenizers_available():
-    # Fast tokenizers
+try:
+    if not is_tokenizers_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    from .utils import dummy_tokenizers_objects
+
+    _import_structure["utils.dummy_tokenizers_objects"] = [
+        name for name in dir(dummy_tokenizers_objects) if not name.startswith("_")
+    ]
+else:
+    # Fast tokenizers structure
     _import_structure["models.albert"].append("AlbertTokenizerFast")
     _import_structure["models.bart"].append("BartTokenizerFast")
     _import_structure["models.barthez"].append("BarthezTokenizerFast")
@@ -498,43 +511,55 @@ if is_tokenizers_available():
     _import_structure["models.xlnet"].append("XLNetTokenizerFast")
     _import_structure["tokenization_utils_fast"] = ["PreTrainedTokenizerFast"]
 
-else:
-    from .utils import dummy_tokenizers_objects
-
-    _import_structure["utils.dummy_tokenizers_objects"] = [
-        name for name in dir(dummy_tokenizers_objects) if not name.startswith("_")
-    ]
 
-if is_sentencepiece_available() and is_tokenizers_available():
-    _import_structure["convert_slow_tokenizer"] = ["SLOW_TO_FAST_CONVERTERS", "convert_slow_tokenizer"]
-else:
+try:
+    if not (is_sentencepiece_available() and is_tokenizers_available()):
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
     from .utils import dummy_sentencepiece_and_tokenizers_objects
 
     _import_structure["utils.dummy_sentencepiece_and_tokenizers_objects"] = [
         name for name in dir(dummy_sentencepiece_and_tokenizers_objects) if not name.startswith("_")
     ]
+else:
+    _import_structure["convert_slow_tokenizer"] = ["SLOW_TO_FAST_CONVERTERS", "convert_slow_tokenizer"]
 
 # Speech-specific objects
-if is_speech_available():
-    _import_structure["models.speech_to_text"].append("Speech2TextFeatureExtractor")
-else:
+try:
+    if not is_speech_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
     from .utils import dummy_speech_objects
 
     _import_structure["utils.dummy_speech_objects"] = [
         name for name in dir(dummy_speech_objects) if not name.startswith("_")
     ]
-
-if is_sentencepiece_available() and is_speech_available():
-    _import_structure["models.speech_to_text"].append("Speech2TextProcessor")
 else:
+    _import_structure["models.speech_to_text"].append("Speech2TextFeatureExtractor")
+
+try:
+    if not (is_sentencepiece_available() and is_speech_available()):
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
     from .utils import dummy_sentencepiece_and_speech_objects
 
     _import_structure["utils.dummy_sentencepiece_and_speech_objects"] = [
         name for name in dir(dummy_sentencepiece_and_speech_objects) if not name.startswith("_")
     ]
+else:
+    _import_structure["models.speech_to_text"].append("Speech2TextProcessor")
 
 # Vision-specific objects
-if is_vision_available():
+try:
+    if not is_vision_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    from .utils import dummy_vision_objects
+
+    _import_structure["utils.dummy_vision_objects"] = [
+        name for name in dir(dummy_vision_objects) if not name.startswith("_")
+    ]
+else:
     _import_structure["image_utils"] = ["ImageFeatureExtractionMixin"]
     _import_structure["models.beit"].append("BeitFeatureExtractor")
     _import_structure["models.clip"].append("CLIPFeatureExtractor")
@@ -556,15 +581,18 @@ if is_vision_available():
     _import_structure["models.vilt"].append("ViltProcessor")
     _import_structure["models.vit"].append("ViTFeatureExtractor")
     _import_structure["models.yolos"].append("YolosFeatureExtractor")
-else:
-    from .utils import dummy_vision_objects
-
-    _import_structure["utils.dummy_vision_objects"] = [
-        name for name in dir(dummy_vision_objects) if not name.startswith("_")
-    ]
 
 # Timm-backed objects
-if is_timm_available() and is_vision_available():
+try:
+    if not (is_timm_available() and is_vision_available()):
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    from .utils import dummy_timm_objects
+
+    _import_structure["utils.dummy_timm_objects"] = [
+        name for name in dir(dummy_timm_objects) if not name.startswith("_")
+    ]
+else:
     _import_structure["models.detr"].extend(
         [
             "DETR_PRETRAINED_MODEL_ARCHIVE_LIST",
@@ -574,14 +602,17 @@ if is_timm_available() and is_vision_available():
             "DetrPreTrainedModel",
         ]
     )
-else:
-    from .utils import dummy_timm_objects
 
-    _import_structure["utils.dummy_timm_objects"] = [
-        name for name in dir(dummy_timm_objects) if not name.startswith("_")
-    ]
+try:
+    if not is_scatter_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    from .utils import dummy_scatter_objects
 
-if is_scatter_available():
+    _import_structure["utils.dummy_scatter_objects"] = [
+        name for name in dir(dummy_scatter_objects) if not name.startswith("_")
+    ]
+else:
     _import_structure["models.tapas"].extend(
         [
             "TAPAS_PRETRAINED_MODEL_ARCHIVE_LIST",
@@ -593,16 +624,17 @@ if is_scatter_available():
             "load_tf_weights_in_tapas",
         ]
     )
-else:
-    from .utils import dummy_scatter_objects
-
-    _import_structure["utils.dummy_scatter_objects"] = [
-        name for name in dir(dummy_scatter_objects) if not name.startswith("_")
-    ]
 
 
 # PyTorch-backed objects
-if is_torch_available():
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    from .utils import dummy_pt_objects
+
+    _import_structure["utils.dummy_pt_objects"] = [name for name in dir(dummy_pt_objects) if not name.startswith("_")]
+else:
     _import_structure["activations"] = []
     _import_structure["benchmark.benchmark"] = ["PyTorchBenchmark"]
     _import_structure["benchmark.benchmark_args"] = ["PyTorchBenchmarkArguments"]
@@ -1723,13 +1755,16 @@ if is_torch_available():
     _import_structure["trainer"] = ["Trainer"]
     _import_structure["trainer_pt_utils"] = ["torch_distributed_zero_first"]
     _import_structure["trainer_seq2seq"] = ["Seq2SeqTrainer"]
-else:
-    from .utils import dummy_pt_objects
-
-    _import_structure["utils.dummy_pt_objects"] = [name for name in dir(dummy_pt_objects) if not name.startswith("_")]
 
 # TensorFlow-backed objects
-if is_tf_available():
+try:
+    if not is_tf_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    from .utils import dummy_tf_objects
+
+    _import_structure["utils.dummy_tf_objects"] = [name for name in dir(dummy_tf_objects) if not name.startswith("_")]
+else:
     _import_structure["activations_tf"] = []
     _import_structure["benchmark.benchmark_args_tf"] = ["TensorFlowBenchmarkArguments"]
     _import_structure["benchmark.benchmark_tf"] = ["TensorFlowBenchmark"]
@@ -2236,13 +2271,18 @@ if is_tf_available():
     _import_structure["tf_utils"] = []
     _import_structure["trainer_tf"] = ["TFTrainer"]
 
-else:
-    from .utils import dummy_tf_objects
-
-    _import_structure["utils.dummy_tf_objects"] = [name for name in dir(dummy_tf_objects) if not name.startswith("_")]
 
 # FLAX-backed objects
-if is_flax_available():
+try:
+    if not is_flax_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    from .utils import dummy_flax_objects
+
+    _import_structure["utils.dummy_flax_objects"] = [
+        name for name in dir(dummy_flax_objects) if not name.startswith("_")
+    ]
+else:
     _import_structure["generation_flax_logits_process"] = [
         "FlaxForcedBOSTokenLogitsProcessor",
         "FlaxForcedEOSTokenLogitsProcessor",
@@ -2469,12 +2509,6 @@ if is_flax_available():
             "FlaxXLMRobertaModel",
         ]
     )
-else:
-    from .utils import dummy_flax_objects
-
-    _import_structure["utils.dummy_flax_objects"] = [
-        name for name in dir(dummy_flax_objects) if not name.startswith("_")
-    ]
 
 
 # Direct imports for type-checking
@@ -2816,7 +2850,12 @@ if TYPE_CHECKING:
         logging,
     )
 
-    if is_sentencepiece_available():
+    try:
+        if not is_sentencepiece_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        from .utils.dummy_sentencepiece_objects import *
+    else:
         from .models.albert import AlbertTokenizer
         from .models.barthez import BarthezTokenizer
         from .models.bartpho import BartphoTokenizer
@@ -2842,10 +2881,14 @@ if TYPE_CHECKING:
         from .models.xlm_prophetnet import XLMProphetNetTokenizer
         from .models.xlm_roberta import XLMRobertaTokenizer
         from .models.xlnet import XLNetTokenizer
-    else:
-        from .utils.dummy_sentencepiece_objects import *
 
-    if is_tokenizers_available():
+    try:
+        if not is_tokenizers_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        from .utils.dummy_tokenizers_objects import *
+    else:
+        # Fast tokenizers imports
         from .models.albert import AlbertTokenizerFast
         from .models.bart import BartTokenizerFast
         from .models.barthez import BarthezTokenizerFast
@@ -2893,25 +2936,36 @@ if TYPE_CHECKING:
         from .models.xlnet import XLNetTokenizerFast
         from .tokenization_utils_fast import PreTrainedTokenizerFast
 
+    try:
+        if not (is_sentencepiece_available() and is_tokenizers_available()):
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        from .utils.dummies_sentencepiece_and_tokenizers_objects import *
     else:
-        from .utils.dummy_tokenizers_objects import *
-
-    if is_sentencepiece_available() and is_tokenizers_available():
         from .convert_slow_tokenizer import SLOW_TO_FAST_CONVERTERS, convert_slow_tokenizer
-    else:
-        from .utils.dummies_sentencepiece_and_tokenizers_objects import *
 
-    if is_speech_available():
-        from .models.speech_to_text import Speech2TextFeatureExtractor
-    else:
+    try:
+        if not is_speech_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
         from .utils.dummy_speech_objects import *
-
-    if is_speech_available() and is_sentencepiece_available():
-        from .models.speech_to_text import Speech2TextProcessor
     else:
+        from .models.speech_to_text import Speech2TextFeatureExtractor
+
+    try:
+        if not (is_speech_available() and is_sentencepiece_available()):
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
         from .utils.dummy_sentencepiece_and_speech_objects import *
+    else:
+        from .models.speech_to_text import Speech2TextProcessor
 
-    if is_vision_available():
+    try:
+        if not is_vision_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        from .utils.dummy_vision_objects import *
+    else:
         from .image_utils import ImageFeatureExtractionMixin
         from .models.beit import BeitFeatureExtractor
         from .models.clip import CLIPFeatureExtractor, CLIPProcessor
@@ -2930,11 +2984,14 @@ if TYPE_CHECKING:
         from .models.vilt import ViltFeatureExtractor, ViltProcessor
         from .models.vit import ViTFeatureExtractor
         from .models.yolos import YolosFeatureExtractor
-    else:
-        from .utils.dummy_vision_objects import *
 
     # Modeling
-    if is_timm_available() and is_vision_available():
+    try:
+        if not (is_timm_available() and is_vision_available()):
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        from .utils.dummy_timm_objects import *
+    else:
         from .models.detr import (
             DETR_PRETRAINED_MODEL_ARCHIVE_LIST,
             DetrForObjectDetection,
@@ -2942,10 +2999,13 @@ if TYPE_CHECKING:
             DetrModel,
             DetrPreTrainedModel,
         )
-    else:
-        from .utils.dummy_timm_objects import *
 
-    if is_scatter_available():
+    try:
+        if not is_scatter_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        from .utils.dummy_scatter_objects import *
+    else:
         from .models.tapas import (
             TAPAS_PRETRAINED_MODEL_ARCHIVE_LIST,
             TapasForMaskedLM,
@@ -2955,10 +3015,13 @@ if TYPE_CHECKING:
             TapasPreTrainedModel,
             load_tf_weights_in_tapas,
         )
-    else:
-        from .utils.dummy_scatter_objects import *
 
-    if is_torch_available():
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        from .utils.dummy_pt_objects import *
+    else:
         # Benchmarks
         from .benchmark.benchmark import PyTorchBenchmark
         from .benchmark.benchmark_args import PyTorchBenchmarkArguments
@@ -3005,6 +3068,8 @@ if TYPE_CHECKING:
         )
         from .generation_utils import top_k_top_p_filtering
         from .modeling_utils import PreTrainedModel
+
+        # PyTorch model imports
         from .models.albert import (
             ALBERT_PRETRAINED_MODEL_ARCHIVE_LIST,
             AlbertForMaskedLM,
@@ -3896,12 +3961,16 @@ if TYPE_CHECKING:
         from .trainer import Trainer
         from .trainer_pt_utils import torch_distributed_zero_first
         from .trainer_seq2seq import Seq2SeqTrainer
-    else:
-        from .utils.dummy_pt_objects import *
 
     # TensorFlow
-    if is_tf_available():
-
+    try:
+        if not is_tf_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        # Import the same objects as dummies to get them in the namespace.
+        # They will raise an import error if the user tries to instantiate / use them.
+        from .utils.dummy_tf_objects import *
+    else:
         from .benchmark.benchmark_args_tf import TensorFlowBenchmarkArguments
 
         # Benchmarks
@@ -3932,6 +4001,8 @@ if TYPE_CHECKING:
             TFLayoutLMPreTrainedModel,
         )
         from .modeling_tf_utils import TFPreTrainedModel, TFSequenceSummary, TFSharedEmbeddings, shape_list
+
+        # TensorFlow model imports
         from .models.albert import (
             TF_ALBERT_PRETRAINED_MODEL_ARCHIVE_LIST,
             TFAlbertForMaskedLM,
@@ -4310,13 +4381,14 @@ if TYPE_CHECKING:
         # Trainer
         from .trainer_tf import TFTrainer
 
-    else:
+    try:
+        if not is_flax_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
         # Import the same objects as dummies to get them in the namespace.
         # They will raise an import error if the user tries to instantiate / use them.
-        from .utils.dummy_tf_objects import *
-
-    if is_flax_available():
-
+        from .utils.dummy_flax_objects import *
+    else:
         from .generation_flax_logits_process import (
             FlaxForcedBOSTokenLogitsProcessor,
             FlaxForcedEOSTokenLogitsProcessor,
@@ -4329,6 +4401,8 @@ if TYPE_CHECKING:
             FlaxTopPLogitsWarper,
         )
         from .modeling_flax_utils import FlaxPreTrainedModel
+
+        # Flax model imports
         from .models.albert import (
             FlaxAlbertForMaskedLM,
             FlaxAlbertForMultipleChoice,
@@ -4494,10 +4568,6 @@ if TYPE_CHECKING:
             FlaxXLMRobertaForTokenClassification,
             FlaxXLMRobertaModel,
         )
-    else:
-        # Import the same objects as dummies to get them in the namespace.
-        # They will raise an import error if the user tries to instantiate / use them.
-        from .utils.dummy_flax_objects import *
 
 else:
     import sys

src/transformers/commands/add_new_model_like.py

@@ -766,7 +766,9 @@ def clean_frameworks_in_init(
         return
 
     remove_pattern = "|".join(to_remove)
-    re_conditional_imports = re.compile(rf"^\s*if is_({remove_pattern})_available\(\):\s*$")
+    re_conditional_imports = re.compile(rf"^\s*if not is_({remove_pattern})_available\(\):\s*$")
+    re_try = re.compile(r"\s*try:")
+    re_else = re.compile(r"\s*else:")
     re_is_xxx_available = re.compile(rf"is_({remove_pattern})_available")
 
     with open(init_file, "r", encoding="utf-8") as f:
@@ -776,10 +778,14 @@ def clean_frameworks_in_init(
     new_lines = []
     idx = 0
     while idx < len(lines):
-        # Conditional imports
-        if re_conditional_imports.search(lines[idx]) is not None:
+        # Conditional imports in try-except-else blocks
+        if (re_conditional_imports.search(lines[idx]) is not None) and (re_try.search(lines[idx - 1]) is not None):
+            # Remove the preceding `try:`
+            new_lines.pop()
+            idx += 1
+            # Iterate until `else:`
+            while is_empty_line(lines[idx]) or re_else.search(lines[idx]) is None:
                 idx += 1
-            while is_empty_line(lines[idx]):
             idx += 1
             indent = find_indent(lines[idx])
             while find_indent(lines[idx]) >= indent or is_empty_line(lines[idx]):
@@ -790,6 +796,7 @@ def clean_frameworks_in_init(
             for framework in to_remove:
                 line = line.replace(f", is_{framework}_available", "")
                 line = line.replace(f"is_{framework}_available, ", "")
+                line = line.replace(f"is_{framework}_available,", "")
                 line = line.replace(f"is_{framework}_available", "")
 
             if len(line.strip()) > 0:
@@ -836,11 +843,11 @@ def add_model_to_main_init(
     while idx < len(lines):
         if not is_empty_line(lines[idx]) and find_indent(lines[idx]) == 0:
             framework = None
-        elif lines[idx].lstrip().startswith("if is_torch_available"):
+        elif lines[idx].lstrip().startswith("if not is_torch_available"):
             framework = "pt"
-        elif lines[idx].lstrip().startswith("if is_tf_available"):
+        elif lines[idx].lstrip().startswith("if not is_tf_available"):
             framework = "tf"
-        elif lines[idx].lstrip().startswith("if is_flax_available"):
+        elif lines[idx].lstrip().startswith("if not is_flax_available"):
             framework = "flax"
 
         # Skip if we are in a framework not wanted.

src/transformers/models/albert/__init__.py

@@ -19,6 +19,7 @@
 from typing import TYPE_CHECKING
 
 from ...utils import (
+    OptionalDependencyNotAvailable,
     _LazyModule,
     is_flax_available,
     is_sentencepiece_available,
@@ -32,13 +33,28 @@ _import_structure = {
     "configuration_albert": ["ALBERT_PRETRAINED_CONFIG_ARCHIVE_MAP", "AlbertConfig", "AlbertOnnxConfig"],
 }
 
-if is_sentencepiece_available():
+try:
+    if not is_sentencepiece_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["tokenization_albert"] = ["AlbertTokenizer"]
 
-if is_tokenizers_available():
+try:
+    if not is_tokenizers_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["tokenization_albert_fast"] = ["AlbertTokenizerFast"]
 
-if is_torch_available():
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_albert"] = [
         "ALBERT_PRETRAINED_MODEL_ARCHIVE_LIST",
         "AlbertForMaskedLM",
@@ -52,7 +68,12 @@ if is_torch_available():
         "load_tf_weights_in_albert",
     ]
 
-if is_tf_available():
+try:
+    if not is_tf_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_tf_albert"] = [
         "TF_ALBERT_PRETRAINED_MODEL_ARCHIVE_LIST",
         "TFAlbertForMaskedLM",
@@ -66,7 +87,12 @@ if is_tf_available():
         "TFAlbertPreTrainedModel",
     ]
 
-if is_flax_available():
+try:
+    if not is_flax_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_flax_albert"] = [
         "FlaxAlbertForMaskedLM",
         "FlaxAlbertForMultipleChoice",
@@ -81,13 +107,28 @@ if is_flax_available():
 if TYPE_CHECKING:
     from .configuration_albert import ALBERT_PRETRAINED_CONFIG_ARCHIVE_MAP, AlbertConfig, AlbertOnnxConfig
 
-    if is_sentencepiece_available():
+    try:
+        if not is_sentencepiece_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .tokenization_albert import AlbertTokenizer
 
-    if is_tokenizers_available():
+    try:
+        if not is_tokenizers_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .tokenization_albert_fast import AlbertTokenizerFast
 
-    if is_torch_available():
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_albert import (
             ALBERT_PRETRAINED_MODEL_ARCHIVE_LIST,
             AlbertForMaskedLM,
@@ -101,7 +142,12 @@ if TYPE_CHECKING:
             load_tf_weights_in_albert,
         )
 
-    if is_tf_available():
+    try:
+        if not is_tf_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_tf_albert import (
             TF_ALBERT_PRETRAINED_MODEL_ARCHIVE_LIST,
             TFAlbertForMaskedLM,
@@ -115,7 +161,12 @@ if TYPE_CHECKING:
             TFAlbertPreTrainedModel,
         )
 
-    if is_flax_available():
+    try:
+        if not is_flax_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_flax_albert import (
             FlaxAlbertForMaskedLM,
             FlaxAlbertForMultipleChoice,

src/transformers/models/auto/__init__.py

@@ -18,7 +18,13 @@
 
 from typing import TYPE_CHECKING
 
-from ...utils import _LazyModule, is_flax_available, is_tf_available, is_torch_available
+from ...utils import (
+    OptionalDependencyNotAvailable,
+    _LazyModule,
+    is_flax_available,
+    is_tf_available,
+    is_torch_available,
+)
 
 
 _import_structure = {
@@ -29,7 +35,12 @@ _import_structure = {
     "tokenization_auto": ["TOKENIZER_MAPPING", "AutoTokenizer"],
 }
 
-if is_torch_available():
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_auto"] = [
         "MODEL_FOR_AUDIO_CLASSIFICATION_MAPPING",
         "MODEL_FOR_AUDIO_XVECTOR_MAPPING",
@@ -81,7 +92,12 @@ if is_torch_available():
         "AutoModelWithLMHead",
     ]
 
-if is_tf_available():
+try:
+    if not is_tf_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_tf_auto"] = [
         "TF_MODEL_FOR_CAUSAL_LM_MAPPING",
         "TF_MODEL_FOR_IMAGE_CLASSIFICATION_MAPPING",
@@ -115,7 +131,12 @@ if is_tf_available():
         "TFAutoModelWithLMHead",
     ]
 
-if is_flax_available():
+try:
+    if not is_flax_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_flax_auto"] = [
         "FLAX_MODEL_FOR_CAUSAL_LM_MAPPING",
         "FLAX_MODEL_FOR_IMAGE_CLASSIFICATION_MAPPING",
@@ -151,7 +172,12 @@ if TYPE_CHECKING:
     from .processing_auto import PROCESSOR_MAPPING, AutoProcessor
     from .tokenization_auto import TOKENIZER_MAPPING, AutoTokenizer
 
-    if is_torch_available():
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_auto import (
             MODEL_FOR_AUDIO_CLASSIFICATION_MAPPING,
             MODEL_FOR_AUDIO_XVECTOR_MAPPING,
@@ -203,7 +229,12 @@ if TYPE_CHECKING:
             AutoModelWithLMHead,
         )
 
-    if is_tf_available():
+    try:
+        if not is_tf_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_tf_auto import (
             TF_MODEL_FOR_CAUSAL_LM_MAPPING,
             TF_MODEL_FOR_IMAGE_CLASSIFICATION_MAPPING,
@@ -237,7 +268,12 @@ if TYPE_CHECKING:
             TFAutoModelWithLMHead,
         )
 
-    if is_flax_available():
+    try:
+        if not is_flax_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_flax_auto import (
             FLAX_MODEL_FOR_CAUSAL_LM_MAPPING,
             FLAX_MODEL_FOR_IMAGE_CLASSIFICATION_MAPPING,

src/transformers/models/bart/__init__.py

@@ -17,7 +17,14 @@
 # limitations under the License.
 from typing import TYPE_CHECKING
 
-from ...utils import _LazyModule, is_flax_available, is_tf_available, is_tokenizers_available, is_torch_available
+from ...utils import (
+    OptionalDependencyNotAvailable,
+    _LazyModule,
+    is_flax_available,
+    is_tf_available,
+    is_tokenizers_available,
+    is_torch_available,
+)
 
 
 _import_structure = {
@@ -25,10 +32,20 @@ _import_structure = {
     "tokenization_bart": ["BartTokenizer"],
 }
 
-if is_tokenizers_available():
+try:
+    if not is_tokenizers_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["tokenization_bart_fast"] = ["BartTokenizerFast"]
 
-if is_torch_available():
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_bart"] = [
         "BART_PRETRAINED_MODEL_ARCHIVE_LIST",
         "BartForCausalLM",
@@ -40,10 +57,20 @@ if is_torch_available():
         "PretrainedBartModel",
     ]
 
-if is_tf_available():
+try:
+    if not is_tf_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_tf_bart"] = ["TFBartForConditionalGeneration", "TFBartModel", "TFBartPretrainedModel"]
 
-if is_flax_available():
+try:
+    if not is_flax_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_flax_bart"] = [
         "FlaxBartDecoderPreTrainedModel",
         "FlaxBartForCausalLM",
@@ -58,10 +85,20 @@ if TYPE_CHECKING:
     from .configuration_bart import BART_PRETRAINED_CONFIG_ARCHIVE_MAP, BartConfig, BartOnnxConfig
     from .tokenization_bart import BartTokenizer
 
-    if is_tokenizers_available():
+    try:
+        if not is_tokenizers_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .tokenization_bart_fast import BartTokenizerFast
 
-    if is_torch_available():
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_bart import (
             BART_PRETRAINED_MODEL_ARCHIVE_LIST,
             BartForCausalLM,
@@ -73,10 +110,20 @@ if TYPE_CHECKING:
             PretrainedBartModel,
         )
 
-    if is_tf_available():
+    try:
+        if not is_tf_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_tf_bart import TFBartForConditionalGeneration, TFBartModel, TFBartPretrainedModel
 
-    if is_flax_available():
+    try:
+        if not is_flax_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_flax_bart import (
             FlaxBartDecoderPreTrainedModel,
             FlaxBartForCausalLM,

src/transformers/models/barthez/__init__.py

@@ -18,24 +18,44 @@
 
 from typing import TYPE_CHECKING
 
-from ...utils import _LazyModule, is_sentencepiece_available, is_tokenizers_available
+from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_sentencepiece_available, is_tokenizers_available
 
 
 _import_structure = {}
 
-if is_sentencepiece_available():
+try:
+    if not is_sentencepiece_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["tokenization_barthez"] = ["BarthezTokenizer"]
 
-if is_tokenizers_available():
+try:
+    if not is_tokenizers_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["tokenization_barthez_fast"] = ["BarthezTokenizerFast"]
 
 
 if TYPE_CHECKING:
 
-    if is_sentencepiece_available():
+    try:
+        if not is_sentencepiece_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .tokenization_barthez import BarthezTokenizer
 
-    if is_tokenizers_available():
+    try:
+        if not is_tokenizers_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .tokenization_barthez_fast import BarthezTokenizerFast
 
 else:

src/transformers/models/bartpho/__init__.py

@@ -18,16 +18,26 @@
 
 from typing import TYPE_CHECKING
 
-from ...utils import _LazyModule, is_sentencepiece_available
+from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_sentencepiece_available
 
 
 _import_structure = {}
 
-if is_sentencepiece_available():
+try:
+    if not is_sentencepiece_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["tokenization_bartpho"] = ["BartphoTokenizer"]
 
 if TYPE_CHECKING:
-    if is_sentencepiece_available():
+    try:
+        if not is_sentencepiece_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .tokenization_bartpho import BartphoTokenizer
 
 else:

src/transformers/models/beit/__init__.py

@@ -18,17 +18,33 @@
 
 from typing import TYPE_CHECKING
 
-from ...utils import _LazyModule, is_flax_available, is_torch_available, is_vision_available
+from ...utils import (
+    OptionalDependencyNotAvailable,
+    _LazyModule,
+    is_flax_available,
+    is_torch_available,
+    is_vision_available,
+)
 
 
 _import_structure = {
     "configuration_beit": ["BEIT_PRETRAINED_CONFIG_ARCHIVE_MAP", "BeitConfig", "BeitOnnxConfig"],
 }
 
-if is_vision_available():
+try:
+    if not is_vision_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["feature_extraction_beit"] = ["BeitFeatureExtractor"]
 
-if is_torch_available():
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_beit"] = [
         "BEIT_PRETRAINED_MODEL_ARCHIVE_LIST",
         "BeitForImageClassification",
@@ -39,7 +55,12 @@ if is_torch_available():
     ]
 
 
-if is_flax_available():
+try:
+    if not is_flax_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_flax_beit"] = [
         "FlaxBeitForImageClassification",
         "FlaxBeitForMaskedImageModeling",
@@ -50,10 +71,20 @@ if is_flax_available():
 if TYPE_CHECKING:
     from .configuration_beit import BEIT_PRETRAINED_CONFIG_ARCHIVE_MAP, BeitConfig, BeitOnnxConfig
 
-    if is_vision_available():
+    try:
+        if not is_vision_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .feature_extraction_beit import BeitFeatureExtractor
 
-    if is_torch_available():
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_beit import (
             BEIT_PRETRAINED_MODEL_ARCHIVE_LIST,
             BeitForImageClassification,
@@ -63,7 +94,12 @@ if TYPE_CHECKING:
             BeitPreTrainedModel,
         )
 
-    if is_flax_available():
+    try:
+        if not is_flax_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_flax_beit import (
             FlaxBeitForImageClassification,
             FlaxBeitForMaskedImageModeling,

src/transformers/models/bert/__init__.py

@@ -18,7 +18,14 @@
 
 from typing import TYPE_CHECKING
 
-from ...utils import _LazyModule, is_flax_available, is_tf_available, is_tokenizers_available, is_torch_available
+from ...utils import (
+    OptionalDependencyNotAvailable,
+    _LazyModule,
+    is_flax_available,
+    is_tf_available,
+    is_tokenizers_available,
+    is_torch_available,
+)
 
 
 _import_structure = {
@@ -26,10 +33,20 @@ _import_structure = {
     "tokenization_bert": ["BasicTokenizer", "BertTokenizer", "WordpieceTokenizer"],
 }
 
-if is_tokenizers_available():
+try:
+    if not is_tokenizers_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["tokenization_bert_fast"] = ["BertTokenizerFast"]
 
-if is_torch_available():
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_bert"] = [
         "BERT_PRETRAINED_MODEL_ARCHIVE_LIST",
         "BertForMaskedLM",
@@ -46,7 +63,12 @@ if is_torch_available():
         "load_tf_weights_in_bert",
     ]
 
-if is_tf_available():
+try:
+    if not is_tf_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_tf_bert"] = [
         "TF_BERT_PRETRAINED_MODEL_ARCHIVE_LIST",
         "TFBertEmbeddings",
@@ -63,7 +85,12 @@ if is_tf_available():
         "TFBertPreTrainedModel",
     ]
 
-if is_flax_available():
+try:
+    if not is_flax_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_flax_bert"] = [
         "FlaxBertForCausalLM",
         "FlaxBertForMaskedLM",
@@ -81,10 +108,20 @@ if TYPE_CHECKING:
     from .configuration_bert import BERT_PRETRAINED_CONFIG_ARCHIVE_MAP, BertConfig, BertOnnxConfig
     from .tokenization_bert import BasicTokenizer, BertTokenizer, WordpieceTokenizer
 
-    if is_tokenizers_available():
+    try:
+        if not is_tokenizers_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .tokenization_bert_fast import BertTokenizerFast
 
-    if is_torch_available():
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_bert import (
             BERT_PRETRAINED_MODEL_ARCHIVE_LIST,
             BertForMaskedLM,
@@ -101,7 +138,12 @@ if TYPE_CHECKING:
             load_tf_weights_in_bert,
         )
 
-    if is_tf_available():
+    try:
+        if not is_tf_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_tf_bert import (
             TF_BERT_PRETRAINED_MODEL_ARCHIVE_LIST,
             TFBertEmbeddings,
@@ -118,7 +160,12 @@ if TYPE_CHECKING:
             TFBertPreTrainedModel,
         )
 
-    if is_flax_available():
+    try:
+        if not is_flax_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_flax_bert import (
             FlaxBertForCausalLM,
             FlaxBertForMaskedLM,

src/transformers/models/bert_generation/__init__.py

@@ -18,17 +18,27 @@
 
 from typing import TYPE_CHECKING
 
-from ...utils import _LazyModule, is_sentencepiece_available, is_torch_available
+from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_sentencepiece_available, is_torch_available
 
 
 _import_structure = {
     "configuration_bert_generation": ["BertGenerationConfig"],
 }
 
-if is_sentencepiece_available():
+try:
+    if not is_sentencepiece_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["tokenization_bert_generation"] = ["BertGenerationTokenizer"]
 
-if is_torch_available():
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_bert_generation"] = [
         "BertGenerationDecoder",
         "BertGenerationEncoder",
@@ -40,10 +50,20 @@ if is_torch_available():
 if TYPE_CHECKING:
     from .configuration_bert_generation import BertGenerationConfig
 
-    if is_sentencepiece_available():
+    try:
+        if not is_sentencepiece_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .tokenization_bert_generation import BertGenerationTokenizer
 
-    if is_torch_available():
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_bert_generation import (
             BertGenerationDecoder,
             BertGenerationEncoder,

src/transformers/models/big_bird/__init__.py

@@ -18,6 +18,7 @@
 from typing import TYPE_CHECKING
 
 from ...utils import (
+    OptionalDependencyNotAvailable,
     _LazyModule,
     is_flax_available,
     is_sentencepiece_available,
@@ -31,13 +32,28 @@ _import_structure = {
     "configuration_big_bird": ["BIG_BIRD_PRETRAINED_CONFIG_ARCHIVE_MAP", "BigBirdConfig", "BigBirdOnnxConfig"],
 }
 
-if is_sentencepiece_available():
+try:
+    if not is_sentencepiece_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["tokenization_big_bird"] = ["BigBirdTokenizer"]
 
-if is_tokenizers_available():
+try:
+    if not is_tokenizers_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["tokenization_big_bird_fast"] = ["BigBirdTokenizerFast"]
 
-if is_torch_available():
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_big_bird"] = [
         "BIG_BIRD_PRETRAINED_MODEL_ARCHIVE_LIST",
         "BigBirdForCausalLM",
@@ -53,7 +69,12 @@ if is_torch_available():
         "load_tf_weights_in_big_bird",
     ]
 
-if is_flax_available():
+try:
+    if not is_flax_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_flax_big_bird"] = [
         "FlaxBigBirdForCausalLM",
         "FlaxBigBirdForMaskedLM",
@@ -69,13 +90,28 @@ if is_flax_available():
 if TYPE_CHECKING:
     from .configuration_big_bird import BIG_BIRD_PRETRAINED_CONFIG_ARCHIVE_MAP, BigBirdConfig, BigBirdOnnxConfig
 
-    if is_sentencepiece_available():
+    try:
+        if not is_sentencepiece_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .tokenization_big_bird import BigBirdTokenizer
 
-    if is_tokenizers_available():
+    try:
+        if not is_tokenizers_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .tokenization_big_bird_fast import BigBirdTokenizerFast
 
-    if is_torch_available():
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_big_bird import (
             BIG_BIRD_PRETRAINED_MODEL_ARCHIVE_LIST,
             BigBirdForCausalLM,
@@ -91,7 +127,12 @@ if TYPE_CHECKING:
             load_tf_weights_in_big_bird,
         )
 
-    if is_flax_available():
+    try:
+        if not is_flax_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_flax_big_bird import (
             FlaxBigBirdForCausalLM,
             FlaxBigBirdForMaskedLM,

src/transformers/models/bigbird_pegasus/__init__.py

@@ -17,7 +17,7 @@
 # limitations under the License.
 from typing import TYPE_CHECKING
 
-from ...utils import _LazyModule, is_torch_available
+from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_torch_available
 
 
 _import_structure = {
@@ -28,7 +28,12 @@ _import_structure = {
     ],
 }
 
-if is_torch_available():
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_bigbird_pegasus"] = [
         "BIGBIRD_PEGASUS_PRETRAINED_MODEL_ARCHIVE_LIST",
         "BigBirdPegasusForCausalLM",
@@ -47,7 +52,12 @@ if TYPE_CHECKING:
         BigBirdPegasusOnnxConfig,
     )
 
-    if is_torch_available():
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_bigbird_pegasus import (
             BIGBIRD_PEGASUS_PRETRAINED_MODEL_ARCHIVE_LIST,
             BigBirdPegasusForCausalLM,

src/transformers/models/blenderbot/__init__.py

@@ -18,7 +18,14 @@
 
 from typing import TYPE_CHECKING
 
-from ...utils import _LazyModule, is_flax_available, is_tf_available, is_tokenizers_available, is_torch_available
+from ...utils import (
+    OptionalDependencyNotAvailable,
+    _LazyModule,
+    is_flax_available,
+    is_tf_available,
+    is_tokenizers_available,
+    is_torch_available,
+)
 
 
 _import_structure = {
@@ -30,10 +37,20 @@ _import_structure = {
     "tokenization_blenderbot": ["BlenderbotTokenizer"],
 }
 
-if is_tokenizers_available():
+try:
+    if not is_tokenizers_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["tokenization_blenderbot_fast"] = ["BlenderbotTokenizerFast"]
 
-if is_torch_available():
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_blenderbot"] = [
         "BLENDERBOT_PRETRAINED_MODEL_ARCHIVE_LIST",
         "BlenderbotForCausalLM",
@@ -43,7 +60,12 @@ if is_torch_available():
     ]
 
 
-if is_tf_available():
+try:
+    if not is_tf_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_tf_blenderbot"] = [
         "TFBlenderbotForConditionalGeneration",
         "TFBlenderbotModel",
@@ -51,7 +73,12 @@ if is_tf_available():
     ]
 
 
-if is_flax_available():
+try:
+    if not is_flax_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_flax_blenderbot"] = [
         "FlaxBlenderbotForConditionalGeneration",
         "FlaxBlenderbotModel",
@@ -67,10 +94,20 @@ if TYPE_CHECKING:
     )
     from .tokenization_blenderbot import BlenderbotTokenizer
 
-    if is_tokenizers_available():
+    try:
+        if not is_tokenizers_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .tokenization_blenderbot_fast import BlenderbotTokenizerFast
 
-    if is_torch_available():
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_blenderbot import (
             BLENDERBOT_PRETRAINED_MODEL_ARCHIVE_LIST,
             BlenderbotForCausalLM,
@@ -79,14 +116,24 @@ if TYPE_CHECKING:
             BlenderbotPreTrainedModel,
         )
 
-    if is_tf_available():
+    try:
+        if not is_tf_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_tf_blenderbot import (
             TFBlenderbotForConditionalGeneration,
             TFBlenderbotModel,
             TFBlenderbotPreTrainedModel,
         )
 
-    if is_flax_available():
+    try:
+        if not is_flax_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_flax_blenderbot import (
             FlaxBlenderbotForConditionalGeneration,
             FlaxBlenderbotModel,

src/transformers/models/blenderbot_small/__init__.py

@@ -17,7 +17,14 @@
 # limitations under the License.
 from typing import TYPE_CHECKING
 
-from ...utils import _LazyModule, is_flax_available, is_tf_available, is_tokenizers_available, is_torch_available
+from ...utils import (
+    OptionalDependencyNotAvailable,
+    _LazyModule,
+    is_flax_available,
+    is_tf_available,
+    is_tokenizers_available,
+    is_torch_available,
+)
 
 
 _import_structure = {
@@ -29,10 +36,20 @@ _import_structure = {
     "tokenization_blenderbot_small": ["BlenderbotSmallTokenizer"],
 }
 
-if is_tokenizers_available():
+try:
+    if not is_tokenizers_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["tokenization_blenderbot_small_fast"] = ["BlenderbotSmallTokenizerFast"]
 
-if is_torch_available():
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_blenderbot_small"] = [
         "BLENDERBOT_SMALL_PRETRAINED_MODEL_ARCHIVE_LIST",
         "BlenderbotSmallForCausalLM",
@@ -41,14 +58,24 @@ if is_torch_available():
         "BlenderbotSmallPreTrainedModel",
     ]
 
-if is_tf_available():
+try:
+    if not is_tf_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_tf_blenderbot_small"] = [
         "TFBlenderbotSmallForConditionalGeneration",
         "TFBlenderbotSmallModel",
         "TFBlenderbotSmallPreTrainedModel",
     ]
 
-if is_flax_available():
+try:
+    if not is_flax_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_flax_blenderbot_small"] = [
         "FlaxBlenderbotSmallForConditionalGeneration",
         "FlaxBlenderbotSmallModel",
@@ -63,10 +90,20 @@ if TYPE_CHECKING:
     )
     from .tokenization_blenderbot_small import BlenderbotSmallTokenizer
 
-    if is_tokenizers_available():
+    try:
+        if not is_tokenizers_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .tokenization_blenderbot_small_fast import BlenderbotSmallTokenizerFast
 
-    if is_torch_available():
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_blenderbot_small import (
             BLENDERBOT_SMALL_PRETRAINED_MODEL_ARCHIVE_LIST,
             BlenderbotSmallForCausalLM,
@@ -75,14 +112,24 @@ if TYPE_CHECKING:
             BlenderbotSmallPreTrainedModel,
         )
 
-    if is_tf_available():
+    try:
+        if not is_tf_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_tf_blenderbot_small import (
             TFBlenderbotSmallForConditionalGeneration,
             TFBlenderbotSmallModel,
             TFBlenderbotSmallPreTrainedModel,
         )
 
-    if is_flax_available():
+    try:
+        if not is_flax_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_flax_blenderbot_small import (
             FlaxBlenderbotSmallForConditionalGeneration,
             FlaxBlenderbotSmallModel,

src/transformers/models/camembert/__init__.py

@@ -19,6 +19,7 @@
 from typing import TYPE_CHECKING
 
 from ...utils import (
+    OptionalDependencyNotAvailable,
     _LazyModule,
     is_sentencepiece_available,
     is_tf_available,
@@ -31,13 +32,28 @@ _import_structure = {
     "configuration_camembert": ["CAMEMBERT_PRETRAINED_CONFIG_ARCHIVE_MAP", "CamembertConfig", "CamembertOnnxConfig"],
 }
 
-if is_sentencepiece_available():
+try:
+    if not is_sentencepiece_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["tokenization_camembert"] = ["CamembertTokenizer"]
 
-if is_tokenizers_available():
+try:
+    if not is_tokenizers_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["tokenization_camembert_fast"] = ["CamembertTokenizerFast"]
 
-if is_torch_available():
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_camembert"] = [
         "CAMEMBERT_PRETRAINED_MODEL_ARCHIVE_LIST",
         "CamembertForCausalLM",
@@ -49,7 +65,12 @@ if is_torch_available():
         "CamembertModel",
     ]
 
-if is_tf_available():
+try:
+    if not is_tf_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_tf_camembert"] = [
         "TF_CAMEMBERT_PRETRAINED_MODEL_ARCHIVE_LIST",
         "TFCamembertForCausalLM",
@@ -65,13 +86,28 @@ if is_tf_available():
 if TYPE_CHECKING:
     from .configuration_camembert import CAMEMBERT_PRETRAINED_CONFIG_ARCHIVE_MAP, CamembertConfig, CamembertOnnxConfig
 
-    if is_sentencepiece_available():
+    try:
+        if not is_sentencepiece_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .tokenization_camembert import CamembertTokenizer
 
-    if is_tokenizers_available():
+    try:
+        if not is_tokenizers_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .tokenization_camembert_fast import CamembertTokenizerFast
 
-    if is_torch_available():
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_camembert import (
             CAMEMBERT_PRETRAINED_MODEL_ARCHIVE_LIST,
             CamembertForCausalLM,
@@ -83,7 +119,12 @@ if TYPE_CHECKING:
             CamembertModel,
         )
 
-    if is_tf_available():
+    try:
+        if not is_tf_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_tf_camembert import (
             TF_CAMEMBERT_PRETRAINED_MODEL_ARCHIVE_LIST,
             TFCamembertForCausalLM,

src/transformers/models/canine/__init__.py

@@ -17,7 +17,7 @@
 # limitations under the License.
 from typing import TYPE_CHECKING
 
-from ...utils import _LazyModule, is_tokenizers_available, is_torch_available
+from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_tokenizers_available, is_torch_available
 
 
 _import_structure = {
@@ -25,7 +25,12 @@ _import_structure = {
     "tokenization_canine": ["CanineTokenizer"],
 }
 
-if is_torch_available():
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_canine"] = [
         "CANINE_PRETRAINED_MODEL_ARCHIVE_LIST",
         "CanineForMultipleChoice",
@@ -43,7 +48,12 @@ if TYPE_CHECKING:
     from .configuration_canine import CANINE_PRETRAINED_CONFIG_ARCHIVE_MAP, CanineConfig
     from .tokenization_canine import CanineTokenizer
 
-    if is_torch_available():
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_canine import (
             CANINE_PRETRAINED_MODEL_ARCHIVE_LIST,
             CanineForMultipleChoice,

src/transformers/models/clip/__init__.py

@@ -18,6 +18,7 @@
 from typing import TYPE_CHECKING
 
 from ...utils import (
+    OptionalDependencyNotAvailable,
     _LazyModule,
     is_flax_available,
     is_tf_available,
@@ -32,14 +33,29 @@ _import_structure = {
     "tokenization_clip": ["CLIPTokenizer"],
 }
 
-if is_tokenizers_available():
+try:
+    if not is_tokenizers_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["tokenization_clip_fast"] = ["CLIPTokenizerFast"]
 
-if is_vision_available():
+try:
+    if not is_vision_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["feature_extraction_clip"] = ["CLIPFeatureExtractor"]
     _import_structure["processing_clip"] = ["CLIPProcessor"]
 
-if is_torch_available():
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_clip"] = [
         "CLIP_PRETRAINED_MODEL_ARCHIVE_LIST",
         "CLIPModel",
@@ -48,7 +64,12 @@ if is_torch_available():
         "CLIPVisionModel",
     ]
 
-if is_tf_available():
+try:
+    if not is_tf_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_tf_clip"] = [
         "TF_CLIP_PRETRAINED_MODEL_ARCHIVE_LIST",
         "TFCLIPModel",
@@ -57,7 +78,12 @@ if is_tf_available():
         "TFCLIPVisionModel",
     ]
 
-if is_flax_available():
+try:
+    if not is_flax_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_flax_clip"] = [
         "FlaxCLIPModel",
         "FlaxCLIPPreTrainedModel",
@@ -72,14 +98,29 @@ if TYPE_CHECKING:
     from .configuration_clip import CLIP_PRETRAINED_CONFIG_ARCHIVE_MAP, CLIPConfig, CLIPTextConfig, CLIPVisionConfig
     from .tokenization_clip import CLIPTokenizer
 
-    if is_tokenizers_available():
+    try:
+        if not is_tokenizers_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .tokenization_clip_fast import CLIPTokenizerFast
 
-    if is_vision_available():
+    try:
+        if not is_vision_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .feature_extraction_clip import CLIPFeatureExtractor
         from .processing_clip import CLIPProcessor
 
-    if is_torch_available():
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_clip import (
             CLIP_PRETRAINED_MODEL_ARCHIVE_LIST,
             CLIPModel,
@@ -88,7 +129,12 @@ if TYPE_CHECKING:
             CLIPVisionModel,
         )
 
-    if is_tf_available():
+    try:
+        if not is_tf_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_tf_clip import (
             TF_CLIP_PRETRAINED_MODEL_ARCHIVE_LIST,
             TFCLIPModel,
@@ -97,7 +143,12 @@ if TYPE_CHECKING:
             TFCLIPVisionModel,
         )
 
-    if is_flax_available():
+    try:
+        if not is_flax_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_flax_clip import (
             FlaxCLIPModel,
             FlaxCLIPPreTrainedModel,

src/transformers/models/convbert/__init__.py

@@ -17,7 +17,13 @@
 # limitations under the License.
 from typing import TYPE_CHECKING
 
-from ...utils import _LazyModule, is_tf_available, is_tokenizers_available, is_torch_available
+from ...utils import (
+    OptionalDependencyNotAvailable,
+    _LazyModule,
+    is_tf_available,
+    is_tokenizers_available,
+    is_torch_available,
+)
 
 
 _import_structure = {
@@ -25,10 +31,20 @@ _import_structure = {
     "tokenization_convbert": ["ConvBertTokenizer"],
 }
 
-if is_tokenizers_available():
+try:
+    if not is_tokenizers_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["tokenization_convbert_fast"] = ["ConvBertTokenizerFast"]
 
-if is_torch_available():
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_convbert"] = [
         "CONVBERT_PRETRAINED_MODEL_ARCHIVE_LIST",
         "ConvBertForMaskedLM",
@@ -43,7 +59,12 @@ if is_torch_available():
     ]
 
 
-if is_tf_available():
+try:
+    if not is_tf_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_tf_convbert"] = [
         "TF_CONVBERT_PRETRAINED_MODEL_ARCHIVE_LIST",
         "TFConvBertForMaskedLM",
@@ -61,10 +82,20 @@ if TYPE_CHECKING:
     from .configuration_convbert import CONVBERT_PRETRAINED_CONFIG_ARCHIVE_MAP, ConvBertConfig, ConvBertOnnxConfig
     from .tokenization_convbert import ConvBertTokenizer
 
-    if is_tokenizers_available():
+    try:
+        if not is_tokenizers_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .tokenization_convbert_fast import ConvBertTokenizerFast
 
-    if is_torch_available():
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_convbert import (
             CONVBERT_PRETRAINED_MODEL_ARCHIVE_LIST,
             ConvBertForMaskedLM,
@@ -78,7 +109,12 @@ if TYPE_CHECKING:
             load_tf_weights_in_convbert,
         )
 
-    if is_tf_available():
+    try:
+        if not is_tf_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_tf_convbert import (
             TF_CONVBERT_PRETRAINED_MODEL_ARCHIVE_LIST,
             TFConvBertForMaskedLM,

src/transformers/models/convnext/__init__.py

@@ -18,17 +18,33 @@
 from typing import TYPE_CHECKING
 
 # rely on isort to merge the imports
-from ...utils import _LazyModule, is_tf_available, is_torch_available, is_vision_available
+from ...utils import (
+    OptionalDependencyNotAvailable,
+    _LazyModule,
+    is_tf_available,
+    is_torch_available,
+    is_vision_available,
+)
 
 
 _import_structure = {
     "configuration_convnext": ["CONVNEXT_PRETRAINED_CONFIG_ARCHIVE_MAP", "ConvNextConfig"],
 }
 
-if is_vision_available():
+try:
+    if not is_vision_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["feature_extraction_convnext"] = ["ConvNextFeatureExtractor"]
 
-if is_torch_available():
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_convnext"] = [
         "CONVNEXT_PRETRAINED_MODEL_ARCHIVE_LIST",
         "ConvNextForImageClassification",
@@ -36,7 +52,12 @@ if is_torch_available():
         "ConvNextPreTrainedModel",
     ]
 
-if is_tf_available():
+try:
+    if not is_tf_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_tf_convnext"] = [
         "TFConvNextForImageClassification",
         "TFConvNextModel",
@@ -46,10 +67,20 @@ if is_tf_available():
 if TYPE_CHECKING:
     from .configuration_convnext import CONVNEXT_PRETRAINED_CONFIG_ARCHIVE_MAP, ConvNextConfig
 
-    if is_vision_available():
+    try:
+        if not is_vision_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .feature_extraction_convnext import ConvNextFeatureExtractor
 
-    if is_torch_available():
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_convnext import (
             CONVNEXT_PRETRAINED_MODEL_ARCHIVE_LIST,
             ConvNextForImageClassification,
@@ -57,7 +88,12 @@ if TYPE_CHECKING:
             ConvNextPreTrainedModel,
         )
 
-    if is_tf_available():
+    try:
+        if not is_tf_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_convnext import TFConvNextForImageClassification, TFConvNextModel, TFConvNextPreTrainedModel
 
 

src/transformers/models/cpm/__init__.py

@@ -18,23 +18,43 @@
 
 from typing import TYPE_CHECKING
 
-from ...utils import _LazyModule, is_sentencepiece_available, is_tokenizers_available
+from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_sentencepiece_available, is_tokenizers_available
 
 
 _import_structure = {}
 
-if is_sentencepiece_available():
+try:
+    if not is_sentencepiece_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["tokenization_cpm"] = ["CpmTokenizer"]
 
-if is_tokenizers_available():
+try:
+    if not is_tokenizers_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["tokenization_cpm_fast"] = ["CpmTokenizerFast"]
 
 
 if TYPE_CHECKING:
-    if is_sentencepiece_available():
+    try:
+        if not is_sentencepiece_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .tokenization_cpm import CpmTokenizer
 
-    if is_tokenizers_available():
+    try:
+        if not is_tokenizers_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .tokenization_cpm_fast import CpmTokenizerFast
 
 else: