M2M100 support for ONNX export (#15193)

* Add M2M100 support for ONNX export

* Delete useless imports

* Add M2M100 to tests

* Fix protobuf issue

docs/source/serialization.mdx

@@ -55,6 +55,7 @@ Ready-made configurations include the following architectures:
 - GPT Neo
 - I-BERT
 - LayoutLM
+- M2M100
 - Marian
 - mBART
 - OpenAI GPT-2
@@ -584,12 +585,12 @@ traced_model(tokens_tensor, segments_tensors)
 
 ### Deploying HuggingFace TorchScript models on AWS using the Neuron SDK
 
-AWS introduced the [Amazon EC2 Inf1](https://aws.amazon.com/ec2/instance-types/inf1/) 
-instance family for low cost, high performance machine learning inference in the cloud. 
-The Inf1 instances are powered by the AWS Inferentia chip, a custom-built hardware accelerator, 
-specializing in deep learning inferencing workloads. 
-[AWS Neuron](https://awsdocs-neuron.readthedocs-hosted.com/en/latest/#) 
-is the SDK for Inferentia that supports tracing and optimizing transformers models for 
+AWS introduced the [Amazon EC2 Inf1](https://aws.amazon.com/ec2/instance-types/inf1/)
+instance family for low cost, high performance machine learning inference in the cloud.
+The Inf1 instances are powered by the AWS Inferentia chip, a custom-built hardware accelerator,
+specializing in deep learning inferencing workloads.
+[AWS Neuron](https://awsdocs-neuron.readthedocs-hosted.com/en/latest/#)
+is the SDK for Inferentia that supports tracing and optimizing transformers models for
 deployment on Inf1. The Neuron SDK provides:
 
 
@@ -600,13 +601,13 @@ deployment on Inf1. The Neuron SDK provides:
 
 #### Implications
 
-Transformers Models based on the [BERT (Bidirectional Encoder Representations from Transformers)](https://huggingface.co/docs/transformers/master/model_doc/bert) 
+Transformers Models based on the [BERT (Bidirectional Encoder Representations from Transformers)](https://huggingface.co/docs/transformers/master/model_doc/bert)
 architecture, or its variants such as [distilBERT](https://huggingface.co/docs/transformers/master/model_doc/distilbert)
- and [roBERTa](https://huggingface.co/docs/transformers/master/model_doc/roberta) 
- will run best on Inf1 for non-generative tasks such as Extractive Question Answering, 
+ and [roBERTa](https://huggingface.co/docs/transformers/master/model_doc/roberta)
+ will run best on Inf1 for non-generative tasks such as Extractive Question Answering,
  Sequence Classification, Token Classification. Alternatively, text generation
-tasks can be adapted to run on Inf1, according to this [AWS Neuron MarianMT tutorial](https://awsdocs-neuron.readthedocs-hosted.com/en/latest/src/examples/pytorch/transformers-marianmt.html). 
-More information about models that can be converted out of the box on Inferentia can be 
+tasks can be adapted to run on Inf1, according to this [AWS Neuron MarianMT tutorial](https://awsdocs-neuron.readthedocs-hosted.com/en/latest/src/examples/pytorch/transformers-marianmt.html).
+More information about models that can be converted out of the box on Inferentia can be
 found in the [Model Architecture Fit section of the Neuron documentation](https://awsdocs-neuron.readthedocs-hosted.com/en/latest/neuron-guide/models/models-inferentia.html#models-inferentia).
 
 #### Dependencies
@@ -618,8 +619,8 @@ Using AWS Neuron to convert models requires the following dependencies and envir
 
 #### Converting a Model for AWS Neuron
 
-Using the same script as in [Using TorchScript in Python](https://huggingface.co/docs/transformers/master/en/serialization#using-torchscript-in-python) 
-to trace a "BertModel", you import `torch.neuron` framework extension to access 
+Using the same script as in [Using TorchScript in Python](https://huggingface.co/docs/transformers/master/en/serialization#using-torchscript-in-python)
+to trace a "BertModel", you import `torch.neuron` framework extension to access
 the components of the Neuron SDK through a Python API.
 
 ```python
@@ -643,5 +644,5 @@ torch.neuron.trace(model, [token_tensor, segments_tensors])
 
 This change enables Neuron SDK to trace the model and optimize it to run in Inf1 instances.
 
-To learn more about AWS Neuron SDK features, tools, example tutorials and latest updates, 
+To learn more about AWS Neuron SDK features, tools, example tutorials and latest updates,
 please see the [AWS NeuronSDK documentation](https://awsdocs-neuron.readthedocs-hosted.com/en/latest/index.html).

src/transformers/models/m2m_100/__init__.py

@@ -21,7 +21,7 @@ from ...file_utils import _LazyModule, is_tokenizers_available, is_torch_availab
 
 
 _import_structure = {
-    "configuration_m2m_100": ["M2M_100_PRETRAINED_CONFIG_ARCHIVE_MAP", "M2M100Config"],
+    "configuration_m2m_100": ["M2M_100_PRETRAINED_CONFIG_ARCHIVE_MAP", "M2M100Config", "M2M100OnnxConfig"],
     "tokenization_m2m_100": ["M2M100Tokenizer"],
 }
 
@@ -36,7 +36,7 @@ if is_torch_available():
 
 
 if TYPE_CHECKING:
-    from .configuration_m2m_100 import M2M_100_PRETRAINED_CONFIG_ARCHIVE_MAP, M2M100Config
+    from .configuration_m2m_100 import M2M_100_PRETRAINED_CONFIG_ARCHIVE_MAP, M2M100Config, M2M100OnnxConfig
     from .tokenization_m2m_100 import M2M100Tokenizer
 
     if is_torch_available():

src/transformers/models/m2m_100/configuration_m2m_100.py

@@ -13,8 +13,14 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 """ M2M100 model configuration"""
+from collections import OrderedDict
+from typing import Any, Mapping, Optional
 
+from ... import PreTrainedTokenizer
 from ...configuration_utils import PretrainedConfig
+from ...file_utils import TensorType, is_torch_available
+from ...onnx import OnnxConfig, OnnxSeq2SeqConfigWithPast
+from ...onnx.utils import compute_effective_axis_dimension
 from ...utils import logging
 
 
@@ -153,3 +159,126 @@ class M2M100Config(PretrainedConfig):
             decoder_start_token_id=decoder_start_token_id,
             **kwargs,
         )
+
+
+class M2M100OnnxConfig(OnnxSeq2SeqConfigWithPast):
+    @property
+    def inputs(self) -> Mapping[str, Mapping[int, str]]:
+        common_inputs = OrderedDict(
+            [
+                ("input_ids", {0: "batch", 1: "encoder_sequence"}),
+                ("attention_mask", {0: "batch", 1: "encoder_sequence"}),
+            ]
+        )
+
+        if self.use_past:
+            common_inputs["decoder_input_ids"] = {0: "batch"}
+            common_inputs["decoder_attention_mask"] = {0: "batch", 1: "past_decoder_sequence + sequence"}
+        else:
+            common_inputs["decoder_input_ids"] = {0: "batch", 1: "decoder_sequence"}
+            common_inputs["decoder_attention_mask"] = {0: "batch", 1: "decoder_sequence"}
+
+        if self.use_past:
+            self.fill_with_past_key_values_(common_inputs, direction="inputs")
+        return common_inputs
+
+    # Copied from BartOnnxConfig._generate_dummy_inputs_for_sequence_classification_and_question_answering
+    # A better name would be _generate_dummy_inputs_for_encoder_and_decoder because sequence classification and question
+    # answering are not supported for M2M100, but this name is preserved to be able to check that the copy matches what
+    # was done for BART so that it can be updated if need be.
+    def _generate_dummy_inputs_for_sequence_classification_and_question_answering(
+        self,
+        tokenizer: PreTrainedTokenizer,
+        batch_size: int = -1,
+        seq_length: int = -1,
+        is_pair: bool = False,
+        framework: Optional[TensorType] = None,
+    ) -> Mapping[str, Any]:
+        # Copied from OnnxConfig.generate_dummy_inputs
+        # Did not use super(OnnxConfigWithPast, self).generate_dummy_inputs for code clarity.
+        # If dynamic axis (-1) we forward with a fixed dimension of 2 samples to avoid optimizations made by ONNX
+        batch_size = compute_effective_axis_dimension(
+            batch_size, fixed_dimension=OnnxConfig.DEFAULT_FIXED_BATCH, num_token_to_add=0
+        )
+
+        # If dynamic axis (-1) we forward with a fixed dimension of 8 tokens to avoid optimizations made by ONNX
+        token_to_add = tokenizer.num_special_tokens_to_add(is_pair)
+        seq_length = compute_effective_axis_dimension(
+            seq_length, fixed_dimension=OnnxConfig.DEFAULT_FIXED_SEQUENCE, num_token_to_add=token_to_add
+        )
+
+        # Generate dummy inputs according to compute batch and sequence
+        dummy_input = [" ".join([tokenizer.unk_token]) * seq_length] * batch_size
+        common_inputs = dict(tokenizer(dummy_input, return_tensors=framework))
+        return common_inputs
+
+    # Copied from transformers.models.bart.configuration_bart.BartOnnxConfig._generate_dummy_inputs_for_default_and_seq2seq_lm
+    def _generate_dummy_inputs_for_default_and_seq2seq_lm(
+        self,
+        tokenizer: PreTrainedTokenizer,
+        batch_size: int = -1,
+        seq_length: int = -1,
+        is_pair: bool = False,
+        framework: Optional[TensorType] = None,
+    ) -> Mapping[str, Any]:
+        encoder_inputs = self._generate_dummy_inputs_for_sequence_classification_and_question_answering(
+            tokenizer, batch_size, seq_length, is_pair, framework
+        )
+
+        # Generate decoder inputs
+        decoder_seq_length = seq_length if not self.use_past else 1
+        decoder_inputs = self._generate_dummy_inputs_for_sequence_classification_and_question_answering(
+            tokenizer, batch_size, decoder_seq_length, is_pair, framework
+        )
+        decoder_inputs = {f"decoder_{name}": tensor for name, tensor in decoder_inputs.items()}
+        common_inputs = dict(**encoder_inputs, **decoder_inputs)
+
+        if self.use_past:
+            if not is_torch_available():
+                raise ValueError("Cannot generate dummy past_keys inputs without PyTorch installed.")
+            else:
+                import torch
+            batch, encoder_seq_length = common_inputs["input_ids"].shape
+            decoder_seq_length = common_inputs["decoder_input_ids"].shape[1]
+            num_encoder_attention_heads, num_decoder_attention_heads = self.num_attention_heads
+            encoder_shape = (
+                batch,
+                num_encoder_attention_heads,
+                encoder_seq_length,
+                self._config.hidden_size // num_encoder_attention_heads,
+            )
+            decoder_past_length = decoder_seq_length + 3
+            decoder_shape = (
+                batch,
+                num_decoder_attention_heads,
+                decoder_past_length,
+                self._config.hidden_size // num_decoder_attention_heads,
+            )
+
+            common_inputs["decoder_attention_mask"] = torch.cat(
+                [common_inputs["decoder_attention_mask"], torch.ones(batch, decoder_past_length)], dim=1
+            )
+
+            common_inputs["past_key_values"] = []
+            # If the number of encoder and decoder layers are present in the model configuration, both are considered
+            num_encoder_layers, num_decoder_layers = self.num_layers
+            min_num_layers = min(num_encoder_layers, num_decoder_layers)
+            max_num_layers = max(num_encoder_layers, num_decoder_layers) - min_num_layers
+            remaining_side_name = "encoder" if num_encoder_layers > num_decoder_layers else "decoder"
+
+            for _ in range(min_num_layers):
+                common_inputs["past_key_values"].append(
+                    (
+                        torch.zeros(decoder_shape),
+                        torch.zeros(decoder_shape),
+                        torch.zeros(encoder_shape),
+                        torch.zeros(encoder_shape),
+                    )
+                )
+            # TODO: test this.
+            shape = encoder_shape if remaining_side_name == "encoder" else decoder_shape
+            for _ in range(min_num_layers, max_num_layers):
+                common_inputs["past_key_values"].append((torch.zeros(shape), torch.zeros(shape)))
+        return common_inputs
+
+    generate_dummy_inputs = _generate_dummy_inputs_for_default_and_seq2seq_lm

src/transformers/onnx/convert.py

@@ -117,21 +117,34 @@ def export_pytorch(
 
             # PyTorch deprecated the `enable_onnx_checker` and `use_external_data_format` arguments in v1.11,
             # so we check the torch version for backwards compatibility
-            if parse(torch.__version__) <= parse("1.10.99"):
+            if parse(torch.__version__) < parse("1.10"):
                 # export can work with named args but the dict containing named args
                 # has to be the last element of the args tuple.
-                onnx_export(
-                    model,
-                    (model_inputs,),
-                    f=output.as_posix(),
-                    input_names=list(config.inputs.keys()),
-                    output_names=onnx_outputs,
-                    dynamic_axes={name: axes for name, axes in chain(config.inputs.items(), config.outputs.items())},
-                    do_constant_folding=True,
-                    use_external_data_format=config.use_external_data_format(model.num_parameters()),
-                    enable_onnx_checker=True,
-                    opset_version=opset,
-                )
+                try:
+                    onnx_export(
+                        model,
+                        (model_inputs,),
+                        f=output.as_posix(),
+                        input_names=list(config.inputs.keys()),
+                        output_names=onnx_outputs,
+                        dynamic_axes={
+                            name: axes for name, axes in chain(config.inputs.items(), config.outputs.items())
+                        },
+                        do_constant_folding=True,
+                        use_external_data_format=config.use_external_data_format(model.num_parameters()),
+                        enable_onnx_checker=True,
+                        opset_version=opset,
+                    )
+                except RuntimeError as err:
+                    message = str(err)
+                    if (
+                        message
+                        == "Exporting model exceed maximum protobuf size of 2GB. Please call torch.onnx.export without setting use_external_data_format parameter."
+                    ):
+                        message = "Exporting model exceed maximum protobuf size of 2GB. Please call torch.onnx.export without setting use_external_data_format parameter or try with torch 1.10+."
+                        raise RuntimeError(message)
+                    else:
+                        raise err
             else:
                 onnx_export(
                     model,

src/transformers/onnx/features.py

@@ -12,6 +12,7 @@ from ..models.gpt2 import GPT2OnnxConfig
 from ..models.gpt_neo import GPTNeoOnnxConfig
 from ..models.ibert import IBertOnnxConfig
 from ..models.layoutlm import LayoutLMOnnxConfig
+from ..models.m2m_100 import M2M100OnnxConfig
 from ..models.marian import MarianOnnxConfig
 from ..models.mbart import MBartOnnxConfig
 from ..models.roberta import RobertaOnnxConfig
@@ -184,6 +185,9 @@ class FeaturesManager:
             "causal-lm-with-past",
             onnx_config_cls=MarianOnnxConfig,
         ),
+        "m2m-100": supported_features_mapping(
+            "default", "default-with-past", "seq2seq-lm", "seq2seq-lm-with-past", onnx_config_cls=M2M100OnnxConfig
+        ),
         "roberta": supported_features_mapping(
             "default",
             "masked-lm",

tests/onnx/test_onnx_v2.py

@@ -190,6 +190,7 @@ PYTORCH_EXPORT_SEQ2SEQ_WITH_PAST_MODELS = {
     ("mbart", "sshleifer/tiny-mbart"),
     ("t5", "t5-small"),
     ("marian", "Helsinki-NLP/opus-mt-en-de"),
+    ("m2m-100", "facebook/m2m100_418M"),
 }
 
 TENSORFLOW_EXPORT_DEFAULT_MODELS = {