Squash commits (#17981)

Co-authored-by: Niels Rogge <nielsrogge@Nielss-MacBook-Pro.local>

docs/source/en/serialization.mdx

@@ -90,6 +90,7 @@ Ready-made configurations include the following architectures:
 - XLM
 - XLM-RoBERTa
 - XLM-RoBERTa-XL
+- YOLOS
 
 In the next two sections, we'll show you how to:
 

src/transformers/models/yolos/__init__.py

@@ -20,7 +20,7 @@ from typing import TYPE_CHECKING
 from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_torch_available, is_vision_available
 
 
-_import_structure = {"configuration_yolos": ["YOLOS_PRETRAINED_CONFIG_ARCHIVE_MAP", "YolosConfig"]}
+_import_structure = {"configuration_yolos": ["YOLOS_PRETRAINED_CONFIG_ARCHIVE_MAP", "YolosConfig", "YolosOnnxConfig"]}
 
 try:
     if not is_vision_available():
@@ -45,7 +45,7 @@ else:
 
 
 if TYPE_CHECKING:
-    from .configuration_yolos import YOLOS_PRETRAINED_CONFIG_ARCHIVE_MAP, YolosConfig
+    from .configuration_yolos import YOLOS_PRETRAINED_CONFIG_ARCHIVE_MAP, YolosConfig, YolosOnnxConfig
 
     try:
         if not is_vision_available():

src/transformers/models/yolos/configuration_yolos.py

@@ -14,7 +14,13 @@
 # limitations under the License.
 """ YOLOS model configuration"""
 
+from collections import OrderedDict
+from typing import Mapping
+
+from packaging import version
+
 from ...configuration_utils import PretrainedConfig
+from ...onnx import OnnxConfig
 from ...utils import logging
 
 
@@ -151,3 +157,24 @@ class YolosConfig(PretrainedConfig):
         self.bbox_loss_coefficient = bbox_loss_coefficient
         self.giou_loss_coefficient = giou_loss_coefficient
         self.eos_coefficient = eos_coefficient
+
+
+class YolosOnnxConfig(OnnxConfig):
+
+    torch_onnx_minimum_version = version.parse("1.11")
+
+    @property
+    def inputs(self) -> Mapping[str, Mapping[int, str]]:
+        return OrderedDict(
+            [
+                ("pixel_values", {0: "batch", 1: "num_channels", 2: "height", 3: "width"}),
+            ]
+        )
+
+    @property
+    def atol_for_validation(self) -> float:
+        return 1e-4
+
+    @property
+    def default_onnx_opset(self) -> int:
+        return 12

src/transformers/onnx/features.py

@@ -452,6 +452,11 @@ class FeaturesManager:
             "question-answering",
             onnx_config_cls="models.xlm_roberta.XLMRobertaOnnxConfig",
         ),
+        "yolos": supported_features_mapping(
+            "default",
+            "object-detection",
+            onnx_config_cls="models.yolos.YolosOnnxConfig",
+        ),
     }
 
     AVAILABLE_FEATURES = sorted(reduce(lambda s1, s2: s1 | s2, (v.keys() for v in _SUPPORTED_MODEL_TYPE.values())))

tests/onnx/test_onnx_v2.py

@@ -202,6 +202,7 @@ PYTORCH_EXPORT_MODELS = {
     ("data2vec-text", "facebook/data2vec-text-base"),
     ("perceiver", "deepmind/language-perceiver", ("masked-lm", "sequence-classification")),
     ("perceiver", "deepmind/vision-perceiver-conv", ("image-classification",)),
+    ("yolos", "hustvl/yolos-tiny"),
 }
 
 PYTORCH_EXPORT_WITH_PAST_MODELS = {