[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/models/vit_mae/__init__.py

@@ -17,14 +17,25 @@
 # limitations under the License.
 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 = {
     "configuration_vit_mae": ["VIT_MAE_PRETRAINED_CONFIG_ARCHIVE_MAP", "ViTMAEConfig"],
 }
 
-if is_torch_available():
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_vit_mae"] = [
         "VIT_MAE_PRETRAINED_MODEL_ARCHIVE_LIST",
         "ViTMAEForPreTraining",
@@ -33,7 +44,12 @@ if is_torch_available():
         "ViTMAEPreTrainedModel",
     ]
 
-if is_tf_available():
+try:
+    if not is_tf_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_tf_vit_mae"] = [
         "TFViTMAEForPreTraining",
         "TFViTMAEModel",
@@ -43,7 +59,12 @@ if is_tf_available():
 if TYPE_CHECKING:
     from .configuration_vit_mae import VIT_MAE_PRETRAINED_CONFIG_ARCHIVE_MAP, ViTMAEConfig
 
-    if is_torch_available():
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_vit_mae import (
             VIT_MAE_PRETRAINED_MODEL_ARCHIVE_LIST,
             ViTMAEForPreTraining,
@@ -52,7 +73,12 @@ if TYPE_CHECKING:
             ViTMAEPreTrainedModel,
         )
 
-    if is_tf_available():
+    try:
+        if not is_tf_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_tf_vit_mae import TFViTMAEForPreTraining, TFViTMAEModel, TFViTMAEPreTrainedModel
 
 

src/transformers/models/wav2vec2/__init__.py

@@ -17,7 +17,13 @@
 # limitations under the License.
 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 = {
@@ -28,7 +34,12 @@ _import_structure = {
 }
 
 
-if is_torch_available():
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_wav2vec2"] = [
         "WAV_2_VEC_2_PRETRAINED_MODEL_ARCHIVE_LIST",
         "Wav2Vec2ForAudioFrameClassification",
@@ -41,7 +52,12 @@ if is_torch_available():
         "Wav2Vec2PreTrainedModel",
     ]
 
-if is_tf_available():
+try:
+    if not is_tf_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_tf_wav2vec2"] = [
         "TF_WAV_2_VEC_2_PRETRAINED_MODEL_ARCHIVE_LIST",
         "TFWav2Vec2ForCTC",
@@ -49,7 +65,12 @@ if is_tf_available():
         "TFWav2Vec2PreTrainedModel",
     ]
 
-if is_flax_available():
+try:
+    if not is_flax_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_flax_wav2vec2"] = [
         "FlaxWav2Vec2ForCTC",
         "FlaxWav2Vec2ForPreTraining",
@@ -64,7 +85,12 @@ if TYPE_CHECKING:
     from .processing_wav2vec2 import Wav2Vec2Processor
     from .tokenization_wav2vec2 import Wav2Vec2CTCTokenizer, Wav2Vec2Tokenizer
 
-    if is_torch_available():
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_wav2vec2 import (
             WAV_2_VEC_2_PRETRAINED_MODEL_ARCHIVE_LIST,
             Wav2Vec2ForAudioFrameClassification,
@@ -77,7 +103,12 @@ if TYPE_CHECKING:
             Wav2Vec2PreTrainedModel,
         )
 
-    if is_tf_available():
+    try:
+        if not is_tf_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_tf_wav2vec2 import (
             TF_WAV_2_VEC_2_PRETRAINED_MODEL_ARCHIVE_LIST,
             TFWav2Vec2ForCTC,
@@ -85,7 +116,12 @@ if TYPE_CHECKING:
             TFWav2Vec2PreTrainedModel,
         )
 
-    if is_flax_available():
+    try:
+        if not is_flax_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_tf_wav2vec2 import (
             FlaxWav2Vec2ForCTC,
             FlaxWav2Vec2ForPreTraining,

src/transformers/models/wavlm/__init__.py

@@ -17,14 +17,19 @@
 # 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 = {
     "configuration_wavlm": ["WAVLM_PRETRAINED_CONFIG_ARCHIVE_MAP", "WavLMConfig"],
 }
 
-if is_torch_available():
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_wavlm"] = [
         "WAVLM_PRETRAINED_MODEL_ARCHIVE_LIST",
         "WavLMForAudioFrameClassification",
@@ -38,7 +43,12 @@ if is_torch_available():
 if TYPE_CHECKING:
     from .configuration_wavlm import WAVLM_PRETRAINED_CONFIG_ARCHIVE_MAP, WavLMConfig
 
-    if is_torch_available():
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_wavlm import (
             WAVLM_PRETRAINED_MODEL_ARCHIVE_LIST,
             WavLMForAudioFrameClassification,

src/transformers/models/xglm/__init__.py

@@ -19,6 +19,7 @@ from typing import TYPE_CHECKING
 
 # rely on isort to merge the imports
 from ...utils import (
+    OptionalDependencyNotAvailable,
     _LazyModule,
     is_flax_available,
     is_sentencepiece_available,
@@ -31,13 +32,28 @@ _import_structure = {
     "configuration_xglm": ["XGLM_PRETRAINED_CONFIG_ARCHIVE_MAP", "XGLMConfig"],
 }
 
-if is_sentencepiece_available():
+try:
+    if not is_sentencepiece_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["tokenization_xglm"] = ["XGLMTokenizer"]
 
-if is_tokenizers_available():
+try:
+    if not is_tokenizers_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["tokenization_xglm_fast"] = ["XGLMTokenizerFast"]
 
-if is_torch_available():
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_xglm"] = [
         "XGLM_PRETRAINED_MODEL_ARCHIVE_LIST",
         "XGLMForCausalLM",
@@ -46,7 +62,12 @@ if is_torch_available():
     ]
 
 
-if is_flax_available():
+try:
+    if not is_flax_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_flax_xglm"] = [
         "FlaxXGLMForCausalLM",
         "FlaxXGLMModel",
@@ -57,16 +78,36 @@ if is_flax_available():
 if TYPE_CHECKING:
     from .configuration_xglm import XGLM_PRETRAINED_CONFIG_ARCHIVE_MAP, XGLMConfig
 
-    if is_sentencepiece_available():
+    try:
+        if not is_sentencepiece_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .tokenization_xglm import XGLMTokenizer
 
-    if is_tokenizers_available():
+    try:
+        if not is_tokenizers_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .tokenization_xglm_fast import XGLMTokenizerFast
 
-    if is_torch_available():
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_xglm import XGLM_PRETRAINED_MODEL_ARCHIVE_LIST, XGLMForCausalLM, XGLMModel, XGLMPreTrainedModel
 
-    if is_flax_available():
+    try:
+        if not is_flax_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_flax_xglm import FlaxXGLMForCausalLM, FlaxXGLMModel, FlaxXGLMPreTrainedModel
 
 

src/transformers/models/xlm/__init__.py

@@ -18,7 +18,7 @@
 
 from typing import TYPE_CHECKING
 
-from ...utils import _LazyModule, is_tf_available, is_torch_available
+from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_tf_available, is_torch_available
 
 
 _import_structure = {
@@ -26,7 +26,12 @@ _import_structure = {
     "tokenization_xlm": ["XLMTokenizer"],
 }
 
-if is_torch_available():
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_xlm"] = [
         "XLM_PRETRAINED_MODEL_ARCHIVE_LIST",
         "XLMForMultipleChoice",
@@ -39,7 +44,12 @@ if is_torch_available():
         "XLMWithLMHeadModel",
     ]
 
-if is_tf_available():
+try:
+    if not is_tf_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_tf_xlm"] = [
         "TF_XLM_PRETRAINED_MODEL_ARCHIVE_LIST",
         "TFXLMForMultipleChoice",
@@ -57,7 +67,12 @@ if TYPE_CHECKING:
     from .configuration_xlm import XLM_PRETRAINED_CONFIG_ARCHIVE_MAP, XLMConfig
     from .tokenization_xlm import XLMTokenizer
 
-    if is_torch_available():
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_xlm import (
             XLM_PRETRAINED_MODEL_ARCHIVE_LIST,
             XLMForMultipleChoice,
@@ -70,7 +85,12 @@ if TYPE_CHECKING:
             XLMWithLMHeadModel,
         )
 
-    if is_tf_available():
+    try:
+        if not is_tf_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_tf_xlm import (
             TF_XLM_PRETRAINED_MODEL_ARCHIVE_LIST,
             TFXLMForMultipleChoice,

src/transformers/models/xlm_prophetnet/__init__.py

@@ -17,7 +17,7 @@
 # limitations under the License.
 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 = {
@@ -27,10 +27,20 @@ _import_structure = {
     ],
 }
 
-if is_sentencepiece_available():
+try:
+    if not is_sentencepiece_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["tokenization_xlm_prophetnet"] = ["XLMProphetNetTokenizer"]
 
-if is_torch_available():
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_xlm_prophetnet"] = [
         "XLM_PROPHETNET_PRETRAINED_MODEL_ARCHIVE_LIST",
         "XLMProphetNetDecoder",
@@ -44,10 +54,20 @@ if is_torch_available():
 if TYPE_CHECKING:
     from .configuration_xlm_prophetnet import XLM_PROPHETNET_PRETRAINED_CONFIG_ARCHIVE_MAP, XLMProphetNetConfig
 
-    if is_sentencepiece_available():
+    try:
+        if not is_sentencepiece_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .tokenization_xlm_prophetnet import XLMProphetNetTokenizer
 
-    if is_torch_available():
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_xlm_prophetnet import (
             XLM_PROPHETNET_PRETRAINED_MODEL_ARCHIVE_LIST,
             XLMProphetNetDecoder,

src/transformers/models/xlm_roberta/__init__.py

@@ -19,6 +19,7 @@
 from typing import TYPE_CHECKING
 
 from ...utils import (
+    OptionalDependencyNotAvailable,
     _LazyModule,
     is_flax_available,
     is_sentencepiece_available,
@@ -36,13 +37,28 @@ _import_structure = {
     ],
 }
 
-if is_sentencepiece_available():
+try:
+    if not is_sentencepiece_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["tokenization_xlm_roberta"] = ["XLMRobertaTokenizer"]
 
-if is_tokenizers_available():
+try:
+    if not is_tokenizers_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["tokenization_xlm_roberta_fast"] = ["XLMRobertaTokenizerFast"]
 
-if is_torch_available():
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_xlm_roberta"] = [
         "XLM_ROBERTA_PRETRAINED_MODEL_ARCHIVE_LIST",
         "XLMRobertaForCausalLM",
@@ -54,7 +70,12 @@ if is_torch_available():
         "XLMRobertaModel",
     ]
 
-if is_tf_available():
+try:
+    if not is_tf_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_tf_xlm_roberta"] = [
         "TF_XLM_ROBERTA_PRETRAINED_MODEL_ARCHIVE_LIST",
         "TFXLMRobertaForMaskedLM",
@@ -65,7 +86,12 @@ if is_tf_available():
         "TFXLMRobertaModel",
     ]
 
-if is_flax_available():
+try:
+    if not is_flax_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_flax_xlm_roberta"] = [
         "FlaxXLMRobertaForMaskedLM",
         "FlaxXLMRobertaForMultipleChoice",
@@ -82,13 +108,28 @@ if TYPE_CHECKING:
         XLMRobertaOnnxConfig,
     )
 
-    if is_sentencepiece_available():
+    try:
+        if not is_sentencepiece_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .tokenization_xlm_roberta import XLMRobertaTokenizer
 
-    if is_tokenizers_available():
+    try:
+        if not is_tokenizers_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .tokenization_xlm_roberta_fast import XLMRobertaTokenizerFast
 
-    if is_torch_available():
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_xlm_roberta import (
             XLM_ROBERTA_PRETRAINED_MODEL_ARCHIVE_LIST,
             XLMRobertaForCausalLM,
@@ -100,7 +141,12 @@ if TYPE_CHECKING:
             XLMRobertaModel,
         )
 
-    if is_tf_available():
+    try:
+        if not is_tf_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_tf_xlm_roberta import (
             TF_XLM_ROBERTA_PRETRAINED_MODEL_ARCHIVE_LIST,
             TFXLMRobertaForMaskedLM,
@@ -111,7 +157,12 @@ if TYPE_CHECKING:
             TFXLMRobertaModel,
         )
 
-    if is_flax_available():
+    try:
+        if not is_flax_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_flax_xlm_roberta import (
             FlaxXLMRobertaForMaskedLM,
             FlaxXLMRobertaForMultipleChoice,

src/transformers/models/xlm_roberta_xl/__init__.py

@@ -18,7 +18,7 @@
 
 from typing import TYPE_CHECKING
 
-from ...utils import _LazyModule, is_torch_available
+from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_torch_available
 
 
 _import_structure = {
@@ -29,7 +29,12 @@ _import_structure = {
     ],
 }
 
-if is_torch_available():
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_xlm_roberta_xl"] = [
         "XLM_ROBERTA_XL_PRETRAINED_MODEL_ARCHIVE_LIST",
         "XLMRobertaXLForCausalLM",
@@ -49,7 +54,12 @@ if TYPE_CHECKING:
         XLMRobertaXLOnnxConfig,
     )
 
-    if is_torch_available():
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_xlm_roberta_xl import (
             XLM_ROBERTA_XL_PRETRAINED_MODEL_ARCHIVE_LIST,
             XLMRobertaXLForCausalLM,

src/transformers/models/xlnet/__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_xlnet": ["XLNET_PRETRAINED_CONFIG_ARCHIVE_MAP", "XLNetConfig"],
 }
 
-if is_sentencepiece_available():
+try:
+    if not is_sentencepiece_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["tokenization_xlnet"] = ["XLNetTokenizer"]
 
-if is_tokenizers_available():
+try:
+    if not is_tokenizers_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["tokenization_xlnet_fast"] = ["XLNetTokenizerFast"]
 
-if is_torch_available():
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_xlnet"] = [
         "XLNET_PRETRAINED_MODEL_ARCHIVE_LIST",
         "XLNetForMultipleChoice",
@@ -51,7 +67,12 @@ if is_torch_available():
         "load_tf_weights_in_xlnet",
     ]
 
-if is_tf_available():
+try:
+    if not is_tf_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_tf_xlnet"] = [
         "TF_XLNET_PRETRAINED_MODEL_ARCHIVE_LIST",
         "TFXLNetForMultipleChoice",
@@ -68,13 +89,28 @@ if is_tf_available():
 if TYPE_CHECKING:
     from .configuration_xlnet import XLNET_PRETRAINED_CONFIG_ARCHIVE_MAP, XLNetConfig
 
-    if is_sentencepiece_available():
+    try:
+        if not is_sentencepiece_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .tokenization_xlnet import XLNetTokenizer
 
-    if is_tokenizers_available():
+    try:
+        if not is_tokenizers_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .tokenization_xlnet_fast import XLNetTokenizerFast
 
-    if is_torch_available():
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_xlnet import (
             XLNET_PRETRAINED_MODEL_ARCHIVE_LIST,
             XLNetForMultipleChoice,
@@ -88,7 +124,12 @@ if TYPE_CHECKING:
             load_tf_weights_in_xlnet,
         )
 
-    if is_tf_available():
+    try:
+        if not is_tf_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_tf_xlnet import (
             TF_XLNET_PRETRAINED_MODEL_ARCHIVE_LIST,
             TFXLNetForMultipleChoice,

src/transformers/models/yolos/__init__.py

@@ -17,17 +17,27 @@
 # limitations under the License.
 from typing import TYPE_CHECKING
 
-from ...utils import _LazyModule, is_torch_available, is_vision_available
+from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_torch_available, is_vision_available
 
 
 _import_structure = {
     "configuration_yolos": ["YOLOS_PRETRAINED_CONFIG_ARCHIVE_MAP", "YolosConfig"],
 }
 
-if is_vision_available():
+try:
+    if not is_vision_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["feature_extraction_yolos"] = ["YolosFeatureExtractor"]
 
-if is_torch_available():
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_yolos"] = [
         "YOLOS_PRETRAINED_MODEL_ARCHIVE_LIST",
         "YolosForObjectDetection",
@@ -39,10 +49,20 @@ if is_torch_available():
 if TYPE_CHECKING:
     from .configuration_yolos import YOLOS_PRETRAINED_CONFIG_ARCHIVE_MAP, YolosConfig
 
-    if is_vision_available():
+    try:
+        if not is_vision_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .feature_extraction_yolos import YolosFeatureExtractor
 
-    if is_torch_available():
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_yolos import (
             YOLOS_PRETRAINED_MODEL_ARCHIVE_LIST,
             YolosForObjectDetection,

src/transformers/models/yoso/__init__.py

@@ -18,14 +18,19 @@
 from typing import TYPE_CHECKING
 
 # rely on isort to merge the imports
-from ...utils import _LazyModule, is_tokenizers_available, is_torch_available
+from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_tokenizers_available, is_torch_available
 
 
 _import_structure = {
     "configuration_yoso": ["YOSO_PRETRAINED_CONFIG_ARCHIVE_MAP", "YosoConfig"],
 }
 
-if is_torch_available():
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_yoso"] = [
         "YOSO_PRETRAINED_MODEL_ARCHIVE_LIST",
         "YosoForMaskedLM",
@@ -42,7 +47,12 @@ if is_torch_available():
 if TYPE_CHECKING:
     from .configuration_yoso import YOSO_PRETRAINED_CONFIG_ARCHIVE_MAP, YosoConfig
 
-    if is_torch_available():
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_yoso import (
             YOSO_PRETRAINED_MODEL_ARCHIVE_LIST,
             YosoForMaskedLM,

src/transformers/utils/__init__.py

@@ -83,6 +83,7 @@ from .import_utils import (
     USE_TF,
     USE_TORCH,
     DummyObject,
+    OptionalDependencyNotAvailable,
     _LazyModule,
     is_apex_available,
     is_bitsandbytes_available,

src/transformers/utils/import_utils.py

@@ -866,3 +866,7 @@ class _LazyModule(ModuleType):
 
     def __reduce__(self):
         return (self.__class__, (self._name, self.__file__, self._import_structure))
+
+
+class OptionalDependencyNotAvailable(BaseException):
+    """Internally used error class for signalling an optional dependency was not found."""

templates/adding_a_new_model/cookiecutter-template-{{cookiecutter.modelname}}/__init__.py

@@ -18,15 +18,23 @@
 from typing import TYPE_CHECKING
 
 # rely on isort to merge the imports
-from ...utils import _LazyModule, is_tokenizers_available
+from ...utils import  _LazyModule, OptionalDependencyNotAvailable, is_tokenizers_available
+
+
 {%- if "TensorFlow" in cookiecutter.generate_tensorflow_pytorch_and_flax %}
 from ...utils import is_tf_available
+
+
 {% endif %}
 {%- if "PyTorch" in cookiecutter.generate_tensorflow_pytorch_and_flax %}
 from ...utils import is_torch_available
+
+
 {% endif %}
 {%- if "Flax" in cookiecutter.generate_tensorflow_pytorch_and_flax %}
 from ...utils import is_flax_available
+
+
 {% endif %}
 
 _import_structure = {
@@ -34,12 +42,22 @@ _import_structure = {
     "tokenization_{{cookiecutter.lowercase_modelname}}": ["{{cookiecutter.camelcase_modelname}}Tokenizer"],
 }
 
-if is_tokenizers_available():
+try:
+    if not is_tokenizers_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["tokenization_{{cookiecutter.lowercase_modelname}}_fast"] = ["{{cookiecutter.camelcase_modelname}}TokenizerFast"]
 
 {%- if "PyTorch" in cookiecutter.generate_tensorflow_pytorch_and_flax %}
 {% if cookiecutter.is_encoder_decoder_model == "False" %}
-if is_torch_available():
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_{{cookiecutter.lowercase_modelname}}"] = [
         "{{cookiecutter.uppercase_modelname}}_PRETRAINED_MODEL_ARCHIVE_LIST",
         "{{cookiecutter.camelcase_modelname}}ForMaskedLM",
@@ -54,7 +72,12 @@ if is_torch_available():
         "load_tf_weights_in_{{cookiecutter.lowercase_modelname}}",
     ]
 {% else %}
-if is_torch_available():
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_{{cookiecutter.lowercase_modelname}}"] = [
         "{{cookiecutter.uppercase_modelname}}_PRETRAINED_MODEL_ARCHIVE_LIST",
         "{{cookiecutter.camelcase_modelname}}ForConditionalGeneration",
@@ -70,7 +93,12 @@ if is_torch_available():
 
 {%- if "TensorFlow" in cookiecutter.generate_tensorflow_pytorch_and_flax %}
 {% if cookiecutter.is_encoder_decoder_model == "False" %}
-if is_tf_available():
+try:
+    if not is_tf_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_tf_{{cookiecutter.lowercase_modelname}}"] = [
         "TF_{{cookiecutter.uppercase_modelname}}_PRETRAINED_MODEL_ARCHIVE_LIST",
         "TF{{cookiecutter.camelcase_modelname}}ForMaskedLM",
@@ -84,7 +112,12 @@ if is_tf_available():
         "TF{{cookiecutter.camelcase_modelname}}PreTrainedModel",
     ]
 {% else %}
-if is_tf_available():
+try:
+    if not is_tf_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_tf_{{cookiecutter.lowercase_modelname}}"] = [
         "TF{{cookiecutter.camelcase_modelname}}ForConditionalGeneration",
         "TF{{cookiecutter.camelcase_modelname}}Model",
@@ -96,7 +129,12 @@ if is_tf_available():
 
 {%- if "Flax" in cookiecutter.generate_tensorflow_pytorch_and_flax %}
 {% if cookiecutter.is_encoder_decoder_model == "False" %}
-if is_flax_available():
+try:
+    if not is_flax_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_flax_{{cookiecutter.lowercase_modelname}}"] = [
         "Flax{{cookiecutter.camelcase_modelname}}ForMaskedLM",
         "Flax{{cookiecutter.camelcase_modelname}}ForCausalLM",
@@ -109,7 +147,12 @@ if is_flax_available():
         "Flax{{cookiecutter.camelcase_modelname}}PreTrainedModel",
     ]
 {% else %}
-if is_flax_available():
+try:
+    if not is_flax_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
     _import_structure["modeling_flax_{{cookiecutter.lowercase_modelname}}"] = [
         "Flax{{cookiecutter.camelcase_modelname}}ForConditionalGeneration",
         "Flax{{cookiecutter.camelcase_modelname}}ForQuestionAnswering",
@@ -125,12 +168,22 @@ if TYPE_CHECKING:
     from .configuration_{{cookiecutter.lowercase_modelname}} import {{cookiecutter.uppercase_modelname}}_PRETRAINED_CONFIG_ARCHIVE_MAP, {{cookiecutter.camelcase_modelname}}Config
     from .tokenization_{{cookiecutter.lowercase_modelname}} import {{cookiecutter.camelcase_modelname}}Tokenizer
 
-    if is_tokenizers_available():
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .tokenization_{{cookiecutter.lowercase_modelname}}_fast import {{cookiecutter.camelcase_modelname}}TokenizerFast
 
 {%- if "PyTorch" in cookiecutter.generate_tensorflow_pytorch_and_flax %}
 {% if cookiecutter.is_encoder_decoder_model == "False" %}
-    if is_torch_available():
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_{{cookiecutter.lowercase_modelname}} import (
             {{cookiecutter.uppercase_modelname}}_PRETRAINED_MODEL_ARCHIVE_LIST,
             {{cookiecutter.camelcase_modelname}}ForMaskedLM,
@@ -145,7 +198,12 @@ if TYPE_CHECKING:
             load_tf_weights_in_{{cookiecutter.lowercase_modelname}},
         )
 {% else %}
-    if is_torch_available():
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_{{cookiecutter.lowercase_modelname}} import (
             {{cookiecutter.uppercase_modelname}}_PRETRAINED_MODEL_ARCHIVE_LIST,
             {{cookiecutter.camelcase_modelname}}ForConditionalGeneration,
@@ -159,7 +217,12 @@ if TYPE_CHECKING:
 {% endif %}
 {%- if "TensorFlow" in cookiecutter.generate_tensorflow_pytorch_and_flax %}
 {% if cookiecutter.is_encoder_decoder_model == "False" %}
-    if is_tf_available():
+    try:
+        if not is_tf_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_tf_{{cookiecutter.lowercase_modelname}} import (
             TF_{{cookiecutter.uppercase_modelname}}_PRETRAINED_MODEL_ARCHIVE_LIST,
             TF{{cookiecutter.camelcase_modelname}}ForMaskedLM,
@@ -173,7 +236,12 @@ if TYPE_CHECKING:
             TF{{cookiecutter.camelcase_modelname}}PreTrainedModel,
         )
 {% else %}
-    if is_tf_available():
+    try:
+        if not is_tf_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_tf_{{cookiecutter.lowercase_modelname}} import (
             TF{{cookiecutter.camelcase_modelname}}ForConditionalGeneration,
             TF{{cookiecutter.camelcase_modelname}}Model,
@@ -183,7 +251,12 @@ if TYPE_CHECKING:
 {% endif %}
 {%- if "Flax" in cookiecutter.generate_tensorflow_pytorch_and_flax %}
 {% if cookiecutter.is_encoder_decoder_model == "False" %}
-    if is_flax_available():
+    try:
+        if not is_flax_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_{{cookiecutter.lowercase_modelname}} import (
             Flax{{cookiecutter.camelcase_modelname}}ForMaskedLM,
             Flax{{cookiecutter.camelcase_modelname}}ForCausalLM,
@@ -196,7 +269,12 @@ if TYPE_CHECKING:
             Flax{{cookiecutter.camelcase_modelname}}PreTrainedModel,
         )
 {% else %}
-    if is_flax_available():
+    try:
+        if not is_flax_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
         from .modeling_{{cookiecutter.lowercase_modelname}} import (
             Flax{{cookiecutter.camelcase_modelname}}ForConditionalGeneration,
             Flax{{cookiecutter.camelcase_modelname}}ForQuestionAnswering,

templates/adding_a_new_model/cookiecutter-template-{{cookiecutter.modelname}}/to_replace_{{cookiecutter.lowercase_modelname}}.py

@@ -115,7 +115,7 @@
 {% endif -%}
 # End.
 
-# Below: "    # Fast tokenizers"
+# Below: "    # Fast tokenizers structure"
 # Replace with:
     _import_structure["models.{{cookiecutter.lowercase_modelname}}"].append("{{cookiecutter.camelcase_modelname}}TokenizerFast")
 # End.
@@ -126,7 +126,7 @@
 # End.
 
 # To replace in: "src/transformers/__init__.py"
-# Below: "    if is_torch_available():" if generating PyTorch
+# Below: "        # PyTorch model imports" if generating PyTorch
 # Replace with:
 {% if cookiecutter.is_encoder_decoder_model == "False" %}
         from .models.{{cookiecutter.lowercase_modelname}} import (
@@ -155,7 +155,7 @@
 {% endif -%}
 # End.
 
-# Below: "    if is_tf_available():" if generating TensorFlow
+# Below: "        # TensorFlow model imports" if generating TensorFlow
 # Replace with:
 {% if cookiecutter.is_encoder_decoder_model == "False" %}
         from .models.{{cookiecutter.lowercase_modelname}} import (
@@ -179,7 +179,7 @@
 {% endif -%}
 # End.
 
-# Below: "    if is_flax_available():" if generating Flax
+# Below: "        # Flax model imports" if generating Flax
 # Replace with:
 {% if cookiecutter.is_encoder_decoder_model == "False" %}
         from .models.{{cookiecutter.lowercase_modelname}} import (
@@ -204,7 +204,7 @@
 {% endif -%}
 # End.
 
-# Below: "    if is_tokenizers_available():"
+# Below: "        # Fast tokenizers imports"
 # Replace with:
         from .models.{{cookiecutter.lowercase_modelname}} import {{cookiecutter.camelcase_modelname}}TokenizerFast
 # End.

utils/check_inits.py

@@ -27,8 +27,8 @@ PATH_TO_TRANSFORMERS = "src/transformers"
 _re_backend = re.compile(r"is\_([a-z_]*)_available()")
 # Catches a line with a key-values pattern: "bla": ["foo", "bar"]
 _re_import_struct_key_value = re.compile(r'\s+"\S*":\s+\[([^\]]*)\]')
-# Catches a line if is_foo_available
-_re_test_backend = re.compile(r"^\s*if\s+is\_[a-z_]*\_available\(\)")
+# Catches a line if not is_foo_available
+_re_test_backend = re.compile(r"^\s*if\s+not\s+is\_[a-z_]*\_available\(\)")
 # Catches a line _import_struct["bla"].append("foo")
 _re_import_struct_add_one = re.compile(r'^\s*_import_structure\["\S*"\]\.append\("(\S*)"\)')
 # Catches a line _import_struct["bla"].extend(["foo", "bar"]) or _import_struct["bla"] = ["foo", "bar"]
@@ -39,6 +39,10 @@ _re_quote_object = re.compile('^\s+"([^"]+)",')
 _re_between_brackets = re.compile("^\s+\[([^\]]+)\]")
 # Catches a line with from foo import bar, bla, boo
 _re_import = re.compile(r"\s+from\s+\S*\s+import\s+([^\(\s].*)\n")
+# Catches a line with try:
+_re_try = re.compile(r"^\s*try:")
+# Catches a line with else:
+_re_else = re.compile(r"^\s*else:")
 
 
 def find_backend(line):
@@ -81,11 +85,21 @@ def parse_init(init_file):
     import_dict_objects = {"none": objects}
     # Let's continue with backend-specific objects in _import_structure
     while not lines[line_index].startswith("if TYPE_CHECKING"):
-        # If the line is an if is_backend_available, we grab all objects associated.
+        # If the line is an if not is_backend_available, we grab all objects associated.
         backend = find_backend(lines[line_index])
+        # Check if the backend declaration is inside a try block:
+        if _re_try.search(lines[line_index - 1]) is None:
+            backend = None
+
         if backend is not None:
             line_index += 1
 
+            # Scroll until we hit the else block of try-except-else
+            while _re_else.search(lines[line_index]) is None:
+                line_index += 1
+
+            line_index += 1
+
             objects = []
             # Until we unindent, add backend objects to the list
             while len(lines[line_index]) <= 1 or lines[line_index].startswith(" " * 4):
@@ -132,9 +146,19 @@ def parse_init(init_file):
     while line_index < len(lines):
         # If the line is an if is_backemd_available, we grab all objects associated.
         backend = find_backend(lines[line_index])
+        # Check if the backend declaration is inside a try block:
+        if _re_try.search(lines[line_index - 1]) is None:
+            backend = None
+
         if backend is not None:
             line_index += 1
 
+            # Scroll until we hit the else block of try-except-else
+            while _re_else.search(lines[line_index]) is None:
+                line_index += 1
+
+            line_index += 1
+
             objects = []
             # Until we unindent, add backend objects to the list
             while len(lines[line_index]) <= 1 or lines[line_index].startswith(" " * 8):