Add RemBERT ONNX config (huggingface#20520)

* rembert onnx config

* formatting

Co-authored-by: Ho <[email protected]>

docs/source/en/serialization.mdx

@@ -93,6 +93,7 @@ Ready-made configurations include the following architectures:
 - OWL-ViT
 - Perceiver
 - PLBart
+- RemBERT
 - ResNet
 - RoBERTa
 - RoFormer

src/transformers/models/rembert/__init__.py

@@ -28,7 +28,9 @@
 )
 
 
-_import_structure = {"configuration_rembert": ["REMBERT_PRETRAINED_CONFIG_ARCHIVE_MAP", "RemBertConfig"]}
+_import_structure = {
+    "configuration_rembert": ["REMBERT_PRETRAINED_CONFIG_ARCHIVE_MAP", "RemBertConfig", "RemBertOnnxConfig"]
+}
 
 try:
     if not is_sentencepiece_available():
@@ -88,7 +90,7 @@
 
 
 if TYPE_CHECKING:
-    from .configuration_rembert import REMBERT_PRETRAINED_CONFIG_ARCHIVE_MAP, RemBertConfig
+    from .configuration_rembert import REMBERT_PRETRAINED_CONFIG_ARCHIVE_MAP, RemBertConfig, RemBertOnnxConfig
 
     try:
         if not is_sentencepiece_available():

src/transformers/models/rembert/configuration_rembert.py

@@ -13,8 +13,11 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 """ RemBERT model configuration"""
+from collections import OrderedDict
+from typing import Mapping
 
 from ...configuration_utils import PretrainedConfig
+from ...onnx import OnnxConfig
 from ...utils import logging
 
 
@@ -135,3 +138,23 @@ def __init__(
         self.layer_norm_eps = layer_norm_eps
         self.use_cache = use_cache
         self.tie_word_embeddings = False
+
+
+class RemBertOnnxConfig(OnnxConfig):
+    @property
+    def inputs(self) -> Mapping[str, Mapping[int, str]]:
+        if self.task == "multiple-choice":
+            dynamic_axis = {0: "batch", 1: "choice", 2: "sequence"}
+        else:
+            dynamic_axis = {0: "batch", 1: "sequence"}
+        return OrderedDict(
+            [
+                ("input_ids", dynamic_axis),
+                ("attention_mask", dynamic_axis),
+                ("token_type_ids", dynamic_axis),
+            ]
+        )
+
+    @property
+    def atol_for_validation(self) -> float:
+        return 1e-4

src/transformers/onnx/features.py

@@ -447,6 +447,16 @@ class FeaturesManager:
             "sequence-classification",
             onnx_config_cls="models.perceiver.PerceiverOnnxConfig",
         ),
+        "rembert": supported_features_mapping(
+            "default",
+            "masked-lm",
+            "causal-lm",
+            "sequence-classification",
+            "multiple-choice",
+            "token-classification",
+            "question-answering",
+            onnx_config_cls="models.rembert.RemBertOnnxConfig",
+        ),
         "resnet": supported_features_mapping(
             "default",
             "image-classification",

tests/onnx/test_onnx_v2.py

@@ -210,6 +210,7 @@ def test_values_override(self):
     ("owlvit", "google/owlvit-base-patch32"),
     ("perceiver", "hf-internal-testing/tiny-random-PerceiverModel", ("masked-lm", "sequence-classification")),
     ("perceiver", "hf-internal-testing/tiny-random-PerceiverModel", ("image-classification",)),
+    ("rembert", "google/rembert"),
     ("resnet", "microsoft/resnet-50"),
     ("roberta", "hf-internal-testing/tiny-random-RobertaModel"),
     ("roformer", "hf-internal-testing/tiny-random-RoFormerModel"),